Employment Data Analysis
The secrets to getting ahead in this company:
1. Work for many years with several different companies before coming to this company. Your Total Time Worked will provide you with a higher job level and higher monthly income.
2. Change Managers frequently. The longer you stay with a manager the longer the time before you will receive you next promotion.
3. Aspire and work toward being a manager or director. These job roles make more money. If you stay with the company long enough and have many years of total work time before coming to the company, you should be able to obtain one of these position.
4. Do not work as a Sales Representative. This job role makes less money and is most likely to quit.
5. Don’t worry about job performance ratings. Everyone receives either an outstanding or an excellent rating. The ratings have not impact on your monthly income.
Executive Summary
Executive Summary
Introduction
This project was created for DDSAnalytics, an analytics company that specializes in talent management solutions for Fortune 1000 companies. According to DDSAnalytics, “Talent management is defined as the iterative process of developing and retaining employees. It may include workforce planning, employee training programs, identifying high-potential employees and reducing/preventing voluntary employee turnover (attrition).” DDSAnalytics has hired our team of data scientists to predict employee turnover by analyzing existing employee data. The company can utilize this analysis to predict factors that lead to employee attrition and potentially prevent voluntary employee attrition.
Project Goals
The main goal of this project is to analyze factors that lead to employee attrition to discover the top most influential factors that lead to attrition. We are also interested in specifics about the company to help DDSAnalytics better understand its employees. We will analyze interesting trends in job roles at the current company as well as job satisfaction. This knowledge can ultimately help DDSAnalytics remain competitive when hiring employees and creating a positive company culture with satisfied employees.
Findings
Attrition Analysis:
Overall, we found that the number one cause of employee attrition is being a Sales Representative. Sales Representatives are 8 times more likely to have attrition than any other category. The second highest cause of attrition is working overtime; employees who work overtime are 5 times more likely to quit. Additionally, people who travel frequently are 4 times more likely to voluntarily leave the company. After these top three factors, the next most influential factors that lead to attrition are: being a Laboratory Technician, being in Human Resources, being a Sales Executive, travelling rarely, being a Research Scientist, being a manager, having a high number of companies worked, and living a far distance from the company.
Job Roles:
Since we are also interested in analyzing the current company, we ran the rest of our analysis only with current employees. We found that the manager is the only job role that overlaps departments and that managers and directors tend to be older and have a higher monthly income than other roles (as expected). The only job role where males do not outnumber females in this company is the Manufacturing Director. The most interesting analysis conducted was on Sales Representatives, since this is the position that is most likely to quit. Sales Representatives have the lowest job levels, the lowest average duration at the company, and change managers the most frequently. All of these factors could contribute to their high turnover rates.
Company Specifics:
Again, we analyzed the company by only including current employees in our analysis. We found that 67% of the company works in Research and Development and that there are 46% more male employees than female employees. We also noticed that job level is strongly correlated to monthly income, which is based on work experience rather than performance. Since everyone in the company receives a performance rating of excellent or outstanding, this holds no weight on income or job satisfaction. Another interesting finding is that the longer employees work for the same manager, there tended to be a longer time there in between promotions. There is a relatively new group of employees at company with 50% of the company working there for less than or equal to 5 years.
Satisfaction:
Generally, job satisfaction decreases as hourly rate decreases, suggesting that hourly rate could be a significant factor leading to satisfaction. Employees with PhD’s, Sales Representatives, and the HR Manager are the most satisfied, which is intriguing given Sales Representatives are the most likely to voluntarily quit. The most dissatisfied employees are the R&D Directors.
Interesting Findings:
There seems to be some discrepancies or errors in the data – we would expect a multiplicative relationship between daily rate and hourly rate or daily rate and monthly rate, but we do not see consistency in this data.
Experimentation:
For our attrition analysis, we ran a logistic regression and fit our model using the STEP procedure as our selection procedure. We ran our model against our test data and came out with 91.8% accuracy in our model, so we sorted the factors that lead to attrition by p-values in descending order of influence. For the remaining analysis on the company, we used linear regression modeling.
Conclusion
This analysis can help DDSAnalytics gain a competitive edge in this industry by creating a positive climate and culture and decreasing the attrition rate. With only 16% attrition, DDSAnalytics does not have a very high attrition rate to begin with. Our recommendations to lower this percentage even further are as follows:
1. Focus time and energy on your Sales Representatives. As this job role was the leading cause of attrition, it would be beneficial to ensure these employees remain satisfied. Things to consider altering about this job role could be the job level the frequent manager changes.
2. Do not grant as many overtime hours – employees are more likely to leave if they work overtime, so limit the number of employees that are working overtime hours.
3. Check in with R&D Directors – this is the most dissatisfied job role in your company.
Please see our report for more detailed findings and experimentation. Please contact Noelle Brown, William Gonzalez, or Ann Nelson with any additional questions.
Data Set
Data Set
## Number of employees in the dataset:## 1470## Number of variables in the dataset:## 35## Number of employees who have left the company and the number if employees who remain:## Left Freq
## No 1233
## Yes 237## 3 Variables had no variation,that is, are constants and will not aid in the analysis## These variables are removed from the dataset.## Number of variables in the dataset, after removing constants:## 32## Count of field types in the dataset:##
## factor integer
## 8 24Job Roles
ggplot(data=as.data.frame(table(attrit$Department,attrit$JobRole,dnn=list("Department","JobRole"))), aes(x= reorder(JobRole,Freq),y=Freq, fill = Department)) +
geom_bar(stat="identity") + coord_flip() + ggtitle("Number of Employees per Job Role by Department") + xlab("") + ylab("")+theme(plot.title=element_text(hjust=0.5))+scale_fill_manual(values=alpha(c("steelblue4", "thistle4", "paleturquoise4")))
The Job Role “Manager” is the only Job Role that is in multiple departments
To aid analysis the values of JobRole for the Managers will be recoded
# Recode Managers
attrit$JobRole <-with(attrit, ifelse(attrit$JobRole == "Manager",paste(attrit$Department,attrit$JobRole), as.character(attrit$JobRole)))
attrit$JobRole <- as.factor(attrit$JobRole)
ggplot(data=as.data.frame(table(attrit$Department,attrit$JobRole,dnn=list("Department","JobRole"))), aes(x= reorder(JobRole,Freq),y=Freq, fill = Department)) +
geom_bar(stat="identity") + coord_flip() + ggtitle("Number of Employees per Job Role by Department \nafter recoding") + xlab("") + ylab("")+theme(plot.title=element_text(hjust=0.5))+scale_fill_manual(values=alpha(c("steelblue4", "thistle4", "paleturquoise4")))
#Recode the variable "JobRole" to a numeric value for modeling
attrit$JRCode[attrit$JobRole=="Healthcare Representative"] <- 1L
attrit$JRCode[attrit$JobRole=="Human Resources"] <- 2L
attrit$JRCode[attrit$JobRole=="Laboratory Technician"] <- 3L
attrit$JRCode[attrit$JobRole=="Manufacturing Director"] <- 4L
attrit$JRCode[attrit$JobRole=="Research Director"] <- 5L
attrit$JRCode[attrit$JobRole=="Sales Executive"] <- 6L
attrit$JRCode[attrit$JobRole=="Research Scientist"] <- 7L
attrit$JRCode[attrit$JobRole=="Sales Representative"] <- 8L
attrit$JRCode[attrit$JobRole=="Human Resources Manager"] <- 9L
attrit$JRCode[attrit$JobRole=="Research & Development Manager"] <- 10L
0.0008## [1] 8e-04#Create a dataframe of current employees only
noattrit <- attrit[attrit$Attrition=="No",]
noattrit$Attrit <- NULLAttrition Analysis
Attrition
First take a quick look at the data. We will name the main dataset used for the attrition portion “attrition”
attrition<-read.csv('Data/CaseStudy2reorder.csv', header = T)
#View(attrition)
head(attrition)## X Attrition BusinessTravel Department EducationField
## 1 1 Yes Travel_Rarely Sales Life Sciences
## 2 2 No Travel_Frequently Research & Development Life Sciences
## 3 3 Yes Travel_Rarely Research & Development Other
## 4 4 No Travel_Frequently Research & Development Life Sciences
## 5 5 No Travel_Rarely Research & Development Medical
## 6 6 No Travel_Frequently Research & Development Life Sciences
## Gender JobRole OverTime MaritalStatus Age DailyRate
## 1 Female Sales Executive Yes Single 41 1102
## 2 Male Research Scientist No Married 49 279
## 3 Male Laboratory Technician Yes Single 37 1373
## 4 Female Research Scientist Yes Married 33 1392
## 5 Male Laboratory Technician No Married 27 591
## 6 Male Laboratory Technician No Single 32 1005
## DistanceFromHome Education EmployeeNumber EnvironmentSatisfaction
## 1 1 2 1 2
## 2 8 1 2 3
## 3 2 2 4 4
## 4 3 4 5 4
## 5 2 1 7 1
## 6 2 2 8 4
## HourlyRate JobInvolvement JobLevel JobSatisfaction MonthlyIncome
## 1 94 3 2 4 5993
## 2 61 2 2 2 5130
## 3 92 2 1 3 2090
## 4 56 3 1 3 2909
## 5 40 3 1 2 3468
## 6 79 3 1 4 3068
## MonthlyRate NumCompaniesWorked PercentSalaryHike PerformanceRating
## 1 19479 8 11 3
## 2 24907 1 23 4
## 3 2396 6 15 3
## 4 23159 1 11 3
## 5 16632 9 12 3
## 6 11864 0 13 3
## RelationshipSatisfaction StockOptionLevel TotalWorkingYears
## 1 1 0 8
## 2 4 1 10
## 3 2 0 7
## 4 3 0 8
## 5 4 1 6
## 6 3 0 8
## TrainingTimesLastYear WorkLifeBalance YearsAtCompany YearsInCurrentRole
## 1 0 1 6 4
## 2 3 3 10 7
## 3 3 3 0 0
## 4 3 3 8 7
## 5 3 3 2 2
## 6 2 2 7 7
## YearsSinceLastPromotion YearsWithCurrManager
## 1 0 5
## 2 1 7
## 3 0 0
## 4 3 0
## 5 2 2
## 6 3 6Drop the first Column, attach dataset for ease of variable recall
attrition<-attrition[,-1] # 1st col obsolete
str(attrition)## 'data.frame': 1470 obs. of 32 variables:
## $ Attrition : Factor w/ 2 levels "No","Yes": 2 1 2 1 1 1 1 1 1 1 ...
## $ BusinessTravel : Factor w/ 3 levels "Non-Travel","Travel_Frequently",..: 3 2 3 2 3 2 3 3 2 3 ...
## $ Department : Factor w/ 3 levels "Human Resources",..: 3 2 2 2 2 2 2 2 2 2 ...
## $ EducationField : Factor w/ 6 levels "Human Resources",..: 2 2 5 2 4 2 4 2 2 4 ...
## $ Gender : Factor w/ 2 levels "Female","Male": 1 2 2 1 2 2 1 2 2 2 ...
## $ JobRole : Factor w/ 9 levels "Healthcare Representative",..: 8 7 3 7 3 3 3 3 5 1 ...
## $ OverTime : Factor w/ 2 levels "No","Yes": 2 1 2 2 1 1 2 1 1 1 ...
## $ MaritalStatus : Factor w/ 3 levels "Divorced","Married",..: 3 2 3 2 2 3 2 1 3 2 ...
## $ Age : int 41 49 37 33 27 32 59 30 38 36 ...
## $ DailyRate : int 1102 279 1373 1392 591 1005 1324 1358 216 1299 ...
## $ DistanceFromHome : int 1 8 2 3 2 2 3 24 23 27 ...
## $ Education : int 2 1 2 4 1 2 3 1 3 3 ...
## $ EmployeeNumber : int 1 2 4 5 7 8 10 11 12 13 ...
## $ EnvironmentSatisfaction : int 2 3 4 4 1 4 3 4 4 3 ...
## $ HourlyRate : int 94 61 92 56 40 79 81 67 44 94 ...
## $ JobInvolvement : int 3 2 2 3 3 3 4 3 2 3 ...
## $ JobLevel : int 2 2 1 1 1 1 1 1 3 2 ...
## $ JobSatisfaction : int 4 2 3 3 2 4 1 3 3 3 ...
## $ MonthlyIncome : int 5993 5130 2090 2909 3468 3068 2670 2693 9526 5237 ...
## $ MonthlyRate : int 19479 24907 2396 23159 16632 11864 9964 13335 8787 16577 ...
## $ NumCompaniesWorked : int 8 1 6 1 9 0 4 1 0 6 ...
## $ PercentSalaryHike : int 11 23 15 11 12 13 20 22 21 13 ...
## $ PerformanceRating : int 3 4 3 3 3 3 4 4 4 3 ...
## $ RelationshipSatisfaction: int 1 4 2 3 4 3 1 2 2 2 ...
## $ StockOptionLevel : int 0 1 0 0 1 0 3 1 0 2 ...
## $ TotalWorkingYears : int 8 10 7 8 6 8 12 1 10 17 ...
## $ TrainingTimesLastYear : int 0 3 3 3 3 2 3 2 2 3 ...
## $ WorkLifeBalance : int 1 3 3 3 3 2 2 3 3 2 ...
## $ YearsAtCompany : int 6 10 0 8 2 7 1 1 9 7 ...
## $ YearsInCurrentRole : int 4 7 0 7 2 7 0 0 7 7 ...
## $ YearsSinceLastPromotion : int 0 1 0 3 2 3 0 0 1 7 ...
## $ YearsWithCurrManager : int 5 7 0 0 2 6 0 0 8 7 ...summary(attrition)## Attrition BusinessTravel Department
## No :1233 Non-Travel : 150 Human Resources : 63
## Yes: 237 Travel_Frequently: 277 Research & Development:961
## Travel_Rarely :1043 Sales :446
##
##
##
##
## EducationField Gender JobRole
## Human Resources : 27 Female:588 Sales Executive :326
## Life Sciences :606 Male :882 Research Scientist :292
## Marketing :159 Laboratory Technician :259
## Medical :464 Manufacturing Director :145
## Other : 82 Healthcare Representative:131
## Technical Degree:132 Manager :102
## (Other) :215
## OverTime MaritalStatus Age DailyRate
## No :1054 Divorced:327 Min. :18.00 Min. : 102.0
## Yes: 416 Married :673 1st Qu.:30.00 1st Qu.: 465.0
## Single :470 Median :36.00 Median : 802.0
## Mean :36.92 Mean : 802.5
## 3rd Qu.:43.00 3rd Qu.:1157.0
## Max. :60.00 Max. :1499.0
##
## DistanceFromHome Education EmployeeNumber EnvironmentSatisfaction
## Min. : 1.000 Min. :1.000 Min. : 1.0 Min. :1.000
## 1st Qu.: 2.000 1st Qu.:2.000 1st Qu.: 491.2 1st Qu.:2.000
## Median : 7.000 Median :3.000 Median :1020.5 Median :3.000
## Mean : 9.193 Mean :2.913 Mean :1024.9 Mean :2.722
## 3rd Qu.:14.000 3rd Qu.:4.000 3rd Qu.:1555.8 3rd Qu.:4.000
## Max. :29.000 Max. :5.000 Max. :2068.0 Max. :4.000
##
## HourlyRate JobInvolvement JobLevel JobSatisfaction
## Min. : 30.00 Min. :1.00 Min. :1.000 Min. :1.000
## 1st Qu.: 48.00 1st Qu.:2.00 1st Qu.:1.000 1st Qu.:2.000
## Median : 66.00 Median :3.00 Median :2.000 Median :3.000
## Mean : 65.89 Mean :2.73 Mean :2.064 Mean :2.729
## 3rd Qu.: 83.75 3rd Qu.:3.00 3rd Qu.:3.000 3rd Qu.:4.000
## Max. :100.00 Max. :4.00 Max. :5.000 Max. :4.000
##
## MonthlyIncome MonthlyRate NumCompaniesWorked PercentSalaryHike
## Min. : 1009 Min. : 2094 Min. :0.000 Min. :11.00
## 1st Qu.: 2911 1st Qu.: 8047 1st Qu.:1.000 1st Qu.:12.00
## Median : 4919 Median :14236 Median :2.000 Median :14.00
## Mean : 6503 Mean :14313 Mean :2.693 Mean :15.21
## 3rd Qu.: 8379 3rd Qu.:20462 3rd Qu.:4.000 3rd Qu.:18.00
## Max. :19999 Max. :26999 Max. :9.000 Max. :25.00
##
## PerformanceRating RelationshipSatisfaction StockOptionLevel
## Min. :3.000 Min. :1.000 Min. :0.0000
## 1st Qu.:3.000 1st Qu.:2.000 1st Qu.:0.0000
## Median :3.000 Median :3.000 Median :1.0000
## Mean :3.154 Mean :2.712 Mean :0.7939
## 3rd Qu.:3.000 3rd Qu.:4.000 3rd Qu.:1.0000
## Max. :4.000 Max. :4.000 Max. :3.0000
##
## TotalWorkingYears TrainingTimesLastYear WorkLifeBalance YearsAtCompany
## Min. : 0.00 Min. :0.000 Min. :1.000 Min. : 0.000
## 1st Qu.: 6.00 1st Qu.:2.000 1st Qu.:2.000 1st Qu.: 3.000
## Median :10.00 Median :3.000 Median :3.000 Median : 5.000
## Mean :11.28 Mean :2.799 Mean :2.761 Mean : 7.008
## 3rd Qu.:15.00 3rd Qu.:3.000 3rd Qu.:3.000 3rd Qu.: 9.000
## Max. :40.00 Max. :6.000 Max. :4.000 Max. :40.000
##
## YearsInCurrentRole YearsSinceLastPromotion YearsWithCurrManager
## Min. : 0.000 Min. : 0.000 Min. : 0.000
## 1st Qu.: 2.000 1st Qu.: 0.000 1st Qu.: 2.000
## Median : 3.000 Median : 1.000 Median : 3.000
## Mean : 4.229 Mean : 2.188 Mean : 4.123
## 3rd Qu.: 7.000 3rd Qu.: 3.000 3rd Qu.: 7.000
## Max. :18.000 Max. :15.000 Max. :17.000
## Some of the categorical variables will not work in an analisys because they are not factors. Convert them to factors.
attrition$Education<-as.factor(attrition$Education)
attrition$EnvironmentSatisfaction<-as.factor(attrition$EnvironmentSatisfaction)
attrition$JobInvolvement<-as.factor(attrition$JobInvolvement)
attrition$JobLevel<-as.factor(attrition$JobLevel)
attrition$JobSatisfaction<-as.factor(attrition$JobSatisfaction)
attrition$PerformanceRating<-as.factor(attrition$PerformanceRating)
attrition$RelationshipSatisfaction<-as.factor(attrition$RelationshipSatisfaction)
attrition$StockOptionLevel<-as.factor(attrition$StockOptionLevel)
attrition$WorkLifeBalance<-as.factor(attrition$WorkLifeBalance) View a Summary of the data
summary(attrition)## Attrition BusinessTravel Department
## No :1233 Non-Travel : 150 Human Resources : 63
## Yes: 237 Travel_Frequently: 277 Research & Development:961
## Travel_Rarely :1043 Sales :446
##
##
##
##
## EducationField Gender JobRole
## Human Resources : 27 Female:588 Sales Executive :326
## Life Sciences :606 Male :882 Research Scientist :292
## Marketing :159 Laboratory Technician :259
## Medical :464 Manufacturing Director :145
## Other : 82 Healthcare Representative:131
## Technical Degree:132 Manager :102
## (Other) :215
## OverTime MaritalStatus Age DailyRate
## No :1054 Divorced:327 Min. :18.00 Min. : 102.0
## Yes: 416 Married :673 1st Qu.:30.00 1st Qu.: 465.0
## Single :470 Median :36.00 Median : 802.0
## Mean :36.92 Mean : 802.5
## 3rd Qu.:43.00 3rd Qu.:1157.0
## Max. :60.00 Max. :1499.0
##
## DistanceFromHome Education EmployeeNumber EnvironmentSatisfaction
## Min. : 1.000 1:170 Min. : 1.0 1:284
## 1st Qu.: 2.000 2:282 1st Qu.: 491.2 2:287
## Median : 7.000 3:572 Median :1020.5 3:453
## Mean : 9.193 4:398 Mean :1024.9 4:446
## 3rd Qu.:14.000 5: 48 3rd Qu.:1555.8
## Max. :29.000 Max. :2068.0
##
## HourlyRate JobInvolvement JobLevel JobSatisfaction MonthlyIncome
## Min. : 30.00 1: 83 1:543 1:289 Min. : 1009
## 1st Qu.: 48.00 2:375 2:534 2:280 1st Qu.: 2911
## Median : 66.00 3:868 3:218 3:442 Median : 4919
## Mean : 65.89 4:144 4:106 4:459 Mean : 6503
## 3rd Qu.: 83.75 5: 69 3rd Qu.: 8379
## Max. :100.00 Max. :19999
##
## MonthlyRate NumCompaniesWorked PercentSalaryHike PerformanceRating
## Min. : 2094 Min. :0.000 Min. :11.00 3:1244
## 1st Qu.: 8047 1st Qu.:1.000 1st Qu.:12.00 4: 226
## Median :14236 Median :2.000 Median :14.00
## Mean :14313 Mean :2.693 Mean :15.21
## 3rd Qu.:20462 3rd Qu.:4.000 3rd Qu.:18.00
## Max. :26999 Max. :9.000 Max. :25.00
##
## RelationshipSatisfaction StockOptionLevel TotalWorkingYears
## 1:276 0:631 Min. : 0.00
## 2:303 1:596 1st Qu.: 6.00
## 3:459 2:158 Median :10.00
## 4:432 3: 85 Mean :11.28
## 3rd Qu.:15.00
## Max. :40.00
##
## TrainingTimesLastYear WorkLifeBalance YearsAtCompany YearsInCurrentRole
## Min. :0.000 1: 80 Min. : 0.000 Min. : 0.000
## 1st Qu.:2.000 2:344 1st Qu.: 3.000 1st Qu.: 2.000
## Median :3.000 3:893 Median : 5.000 Median : 3.000
## Mean :2.799 4:153 Mean : 7.008 Mean : 4.229
## 3rd Qu.:3.000 3rd Qu.: 9.000 3rd Qu.: 7.000
## Max. :6.000 Max. :40.000 Max. :18.000
##
## YearsSinceLastPromotion YearsWithCurrManager
## Min. : 0.000 Min. : 0.000
## 1st Qu.: 0.000 1st Qu.: 2.000
## Median : 1.000 Median : 3.000
## Mean : 2.188 Mean : 4.123
## 3rd Qu.: 3.000 3rd Qu.: 7.000
## Max. :15.000 Max. :17.000
## # attach for ease of running code
attach(attrition)Some initial things to note here:
- Only 16% attrition in the data
- BusinessTravel, OverTime, JobInvolvement, PerformanceRating, seem to be unbalanced. All other Variables seem balanced.
- YearsAtCompany, YearsInCurrentRole, YearsSinceLastPromotion, YearsWithCurrManager may be skewed
Need logistic regression
*** The next several steps are the model fitting steps
Model fitting (Steps 1~zz)
1. Split the data into train and test data
We want to be able to create an appropriate model the test that model on a subset of the data. We will create a “training” dataset and a “test” one.
set.seed(123)
split <- sample.split(attrition$Attrition, SplitRatio = 0.80)
#get training and test data
data.train <- subset(attrition, split == TRUE)
data.test <- subset(attrition, split == FALSE)2. Define null + full models for the training data
We need a NULL model (with no variables) to compare our training model to. We will compare the Deviance and AIC of the NULL Model vs the training model. We are looking for significant drop in both of these stats as a validation that the model is working.
options(width = 300) ## print
null.model.train<-glm(formula=Attrition~1, family = binomial(link="logit"), data=data.train)
full.model.train<-glm(formula=Attrition~., family = binomial(link="logit"), data=data.train)
summary(full.model.train)##
## Call:
## glm(formula = Attrition ~ ., family = binomial(link = "logit"),
## data = data.train)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.8888 -0.4561 -0.1967 -0.0584 3.6739
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -9.397e+00 6.424e+02 -0.015 0.988330
## BusinessTravelTravel_Frequently 1.973e+00 4.971e-01 3.970 7.19e-05 ***
## BusinessTravelTravel_Rarely 9.463e-01 4.612e-01 2.052 0.040162 *
## DepartmentResearch & Development 1.347e+01 6.424e+02 0.021 0.983277
## DepartmentSales 1.347e+01 6.424e+02 0.021 0.983274
## EducationFieldLife Sciences -9.159e-01 9.705e-01 -0.944 0.345297
## EducationFieldMarketing -5.541e-01 1.026e+00 -0.540 0.589208
## EducationFieldMedical -1.061e+00 9.657e-01 -1.099 0.271850
## EducationFieldOther -1.062e+00 1.052e+00 -1.010 0.312496
## EducationFieldTechnical Degree -1.206e-01 9.912e-01 -0.122 0.903150
## GenderMale 4.060e-01 2.172e-01 1.869 0.061656 .
## JobRoleHuman Resources 1.378e+01 6.424e+02 0.021 0.982890
## JobRoleLaboratory Technician 6.647e-01 6.741e-01 0.986 0.324097
## JobRoleManager -1.572e-01 1.220e+00 -0.129 0.897457
## JobRoleManufacturing Director 5.664e-01 6.402e-01 0.885 0.376268
## JobRoleResearch Director -2.135e+00 1.388e+00 -1.538 0.124131
## JobRoleResearch Scientist -6.835e-01 7.058e-01 -0.968 0.332903
## JobRoleSales Executive 1.434e+00 1.450e+00 0.989 0.322653
## JobRoleSales Representative 1.165e+00 1.528e+00 0.763 0.445612
## OverTimeYes 2.066e+00 2.357e-01 8.769 < 2e-16 ***
## MaritalStatusMarried 3.710e-01 3.209e-01 1.156 0.247708
## MaritalStatusSingle 6.425e-01 4.564e-01 1.408 0.159182
## Age -2.812e-02 1.589e-02 -1.770 0.076713 .
## DailyRate -3.854e-04 2.606e-04 -1.479 0.139271
## DistanceFromHome 5.149e-02 1.286e-02 4.003 6.26e-05 ***
## Education2 3.560e-01 3.918e-01 0.909 0.363550
## Education3 2.119e-01 3.502e-01 0.605 0.545108
## Education4 3.073e-01 3.747e-01 0.820 0.412153
## Education5 2.933e-01 7.307e-01 0.401 0.688090
## EmployeeNumber -1.677e-04 1.839e-04 -0.912 0.361818
## EnvironmentSatisfaction2 -1.334e+00 3.360e-01 -3.969 7.22e-05 ***
## EnvironmentSatisfaction3 -1.250e+00 2.980e-01 -4.195 2.73e-05 ***
## EnvironmentSatisfaction4 -1.410e+00 2.990e-01 -4.716 2.40e-06 ***
## HourlyRate 4.870e-03 5.254e-03 0.927 0.354039
## JobInvolvement2 -1.309e+00 4.309e-01 -3.037 0.002388 **
## JobInvolvement3 -1.681e+00 4.089e-01 -4.112 3.93e-05 ***
## JobInvolvement4 -2.041e+00 5.415e-01 -3.769 0.000164 ***
## JobLevel2 -1.677e+00 5.263e-01 -3.186 0.001444 **
## JobLevel3 3.282e-01 8.144e-01 0.403 0.686981
## JobLevel4 -6.879e-01 1.400e+00 -0.491 0.623217
## JobLevel5 1.452e+00 1.975e+00 0.735 0.462083
## JobSatisfaction2 -4.466e-01 3.169e-01 -1.409 0.158823
## JobSatisfaction3 -5.249e-01 2.866e-01 -1.831 0.067051 .
## JobSatisfaction4 -1.214e+00 3.064e-01 -3.961 7.47e-05 ***
## MonthlyIncome -1.526e-04 1.070e-04 -1.425 0.154053
## MonthlyRate 1.950e-05 1.482e-05 1.316 0.188204
## NumCompaniesWorked 2.007e-01 4.632e-02 4.332 1.48e-05 ***
## PercentSalaryHike -3.810e-02 4.673e-02 -0.815 0.414835
## PerformanceRating4 5.542e-02 4.846e-01 0.114 0.908947
## RelationshipSatisfaction2 -8.635e-01 3.411e-01 -2.532 0.011356 *
## RelationshipSatisfaction3 -8.888e-01 3.029e-01 -2.934 0.003344 **
## RelationshipSatisfaction4 -9.020e-01 3.004e-01 -3.003 0.002671 **
## StockOptionLevel1 -1.050e+00 3.627e-01 -2.894 0.003803 **
## StockOptionLevel2 -1.066e+00 4.908e-01 -2.172 0.029866 *
## StockOptionLevel3 -5.810e-01 5.815e-01 -0.999 0.317755
## TotalWorkingYears -6.285e-02 3.547e-02 -1.772 0.076398 .
## TrainingTimesLastYear -1.377e-01 8.333e-02 -1.652 0.098548 .
## WorkLifeBalance2 -1.040e+00 4.259e-01 -2.441 0.014628 *
## WorkLifeBalance3 -1.571e+00 3.999e-01 -3.930 8.51e-05 ***
## WorkLifeBalance4 -1.262e+00 4.845e-01 -2.604 0.009216 **
## YearsAtCompany 1.210e-01 4.956e-02 2.442 0.014605 *
## YearsInCurrentRole -1.620e-01 5.740e-02 -2.823 0.004758 **
## YearsSinceLastPromotion 1.455e-01 5.012e-02 2.904 0.003690 **
## YearsWithCurrManager -1.246e-01 5.428e-02 -2.295 0.021737 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 1040.18 on 1175 degrees of freedom
## Residual deviance: 633.03 on 1112 degrees of freedom
## AIC: 761.03
##
## Number of Fisher Scoring iterations: 15There are a few things to note here:
- We can see the far-right column with asterisks representing the lowest p-values. Since this is the whole model, and it has not been “fitted,” the values here represent individual variables, that, on their own, they each appear to have some significance and explain Attrition.
- The “Estimate” column gies us the Log Odds (logit scale). This will back-transformed in a later step to actual Odds-Ratio, which is the probability that a covariate value influences Attrition in terms of Odds.
- The Null Deviance (1040.18) is a measure of lack of fit of the model without any variables.
- The Residual Deviance (633.03) is a measure of lack of fit of the model with all variables included.
- AIC (761.03) is a measure of how well the model is fitting. We want to keep this as low as possible.
Sorting Model by p-value descending to view important covariates
#sort summary by p-value descending
tempDF<-as.data.frame(summary(full.model.train)$coefficients)
tempDF[order(tempDF$`Pr(>|z|)`),]## Estimate Std. Error z value Pr(>|z|)
## OverTimeYes 2.066496e+00 2.356697e-01 8.76861346 1.808798e-18
## EnvironmentSatisfaction4 -1.410304e+00 2.990337e-01 -4.71620534 2.402839e-06
## NumCompaniesWorked 2.006728e-01 4.632009e-02 4.33230549 1.475560e-05
## EnvironmentSatisfaction3 -1.250120e+00 2.980255e-01 -4.19467519 2.732629e-05
## JobInvolvement3 -1.681390e+00 4.089481e-01 -4.11150059 3.930959e-05
## DistanceFromHome 5.148581e-02 1.286262e-02 4.00274540 6.261167e-05
## BusinessTravelTravel_Frequently 1.973438e+00 4.970795e-01 3.97006641 7.185260e-05
## EnvironmentSatisfaction2 -1.333660e+00 3.360157e-01 -3.96904008 7.216274e-05
## JobSatisfaction4 -1.213550e+00 3.063924e-01 -3.96077056 7.470829e-05
## WorkLifeBalance3 -1.571416e+00 3.998989e-01 -3.92953309 8.511096e-05
## JobInvolvement4 -2.040537e+00 5.414673e-01 -3.76853114 1.642110e-04
## JobLevel2 -1.676527e+00 5.262505e-01 -3.18579714 1.443558e-03
## JobInvolvement2 -1.308709e+00 4.308921e-01 -3.03720794 2.387807e-03
## RelationshipSatisfaction4 -9.020471e-01 3.003511e-01 -3.00330930 2.670609e-03
## RelationshipSatisfaction3 -8.887712e-01 3.029005e-01 -2.93420155 3.344070e-03
## YearsSinceLastPromotion 1.455322e-01 5.012251e-02 2.90352946 3.689823e-03
## StockOptionLevel1 -1.049812e+00 3.627458e-01 -2.89407158 3.802815e-03
## YearsInCurrentRole -1.620440e-01 5.740223e-02 -2.82295655 4.758302e-03
## WorkLifeBalance4 -1.261649e+00 4.845143e-01 -2.60394504 9.215753e-03
## RelationshipSatisfaction2 -8.634599e-01 3.410797e-01 -2.53154875 1.135600e-02
## YearsAtCompany 1.210194e-01 4.955692e-02 2.44202874 1.460498e-02
## WorkLifeBalance2 -1.039934e+00 4.259488e-01 -2.44145443 1.462823e-02
## YearsWithCurrManager -1.245621e-01 5.427698e-02 -2.29493430 2.173689e-02
## StockOptionLevel2 -1.065877e+00 4.907660e-01 -2.17186337 2.986597e-02
## BusinessTravelTravel_Rarely 9.463215e-01 4.611534e-01 2.05207533 4.016234e-02
## GenderMale 4.059671e-01 2.172388e-01 1.86875916 6.165633e-02
## JobSatisfaction3 -5.249475e-01 2.866475e-01 -1.83133497 6.705056e-02
## TotalWorkingYears -6.285103e-02 3.546935e-02 -1.77198141 7.639764e-02
## Age -2.811934e-02 1.588587e-02 -1.77008462 7.671304e-02
## TrainingTimesLastYear -1.376637e-01 8.333496e-02 -1.65193159 9.854850e-02
## JobRoleResearch Director -2.134856e+00 1.388377e+00 -1.53766315 1.241310e-01
## DailyRate -3.853517e-04 2.606352e-04 -1.47850948 1.392715e-01
## MonthlyIncome -1.525736e-04 1.070422e-04 -1.42535949 1.540533e-01
## JobSatisfaction2 -4.465805e-01 3.169395e-01 -1.40904014 1.588233e-01
## MaritalStatusSingle 6.425277e-01 4.563969e-01 1.40782657 1.591824e-01
## MonthlyRate 1.950394e-05 1.482162e-05 1.31591095 1.882039e-01
## MaritalStatusMarried 3.709706e-01 3.209272e-01 1.15593375 2.477083e-01
## EducationFieldMedical -1.061073e+00 9.656534e-01 -1.09881307 2.718496e-01
## EducationFieldOther -1.062406e+00 1.051888e+00 -1.00999854 3.124960e-01
## StockOptionLevel3 -5.809903e-01 5.815237e-01 -0.99908270 3.177546e-01
## JobRoleSales Executive 1.434065e+00 1.449984e+00 0.98902152 3.226526e-01
## JobRoleLaboratory Technician 6.647463e-01 6.741342e-01 0.98607410 3.240968e-01
## JobRoleResearch Scientist -6.834538e-01 7.058412e-01 -0.96828274 3.329032e-01
## EducationFieldLife Sciences -9.159258e-01 9.705168e-01 -0.94375056 3.452971e-01
## HourlyRate 4.869592e-03 5.254288e-03 0.92678425 3.540386e-01
## EmployeeNumber -1.677276e-04 1.839306e-04 -0.91190664 3.618179e-01
## Education2 3.559715e-01 3.917708e-01 0.90862200 3.635497e-01
## JobRoleManufacturing Director 5.664383e-01 6.401926e-01 0.88479366 3.762679e-01
## Education4 3.073242e-01 3.747351e-01 0.82011057 4.121531e-01
## PercentSalaryHike -3.810140e-02 4.672639e-02 -0.81541506 4.148348e-01
## JobRoleSales Representative 1.165268e+00 1.527716e+00 0.76275124 4.456118e-01
## JobLevel5 1.452142e+00 1.974573e+00 0.73542100 4.620831e-01
## Education3 2.119179e-01 3.502153e-01 0.60510745 5.451076e-01
## EducationFieldMarketing -5.540912e-01 1.026125e+00 -0.53998403 5.892080e-01
## JobLevel4 -6.878840e-01 1.400140e+00 -0.49129655 6.232167e-01
## JobLevel3 3.281793e-01 8.144302e-01 0.40295569 6.869808e-01
## Education5 2.933429e-01 7.307101e-01 0.40144909 6.880895e-01
## JobRoleManager -1.572259e-01 1.219999e+00 -0.12887382 8.974575e-01
## EducationFieldTechnical Degree -1.206066e-01 9.911554e-01 -0.12168282 9.031502e-01
## PerformanceRating4 5.542427e-02 4.846185e-01 0.11436679 9.089470e-01
## JobRoleHuman Resources 1.377763e+01 6.424413e+02 0.02144574 9.828901e-01
## DepartmentSales 1.346853e+01 6.424410e+02 0.02096462 9.832739e-01
## DepartmentResearch & Development 1.346568e+01 6.424409e+02 0.02096018 9.832774e-01
## (Intercept) -9.396688e+00 6.424429e+02 -0.01462650 9.883302e-01We can see the top variables, that, on their own, seem to have an effect in Attrition.
3. Run ANOVA
We want to run an ANOVA To further understand the impact of each variable into the model.
anova(full.model.train, test="Chisq")## Analysis of Deviance Table
##
## Model: binomial, link: logit
##
## Response: Attrition
##
## Terms added sequentially (first to last)
##
##
## Df Deviance Resid. Df Resid. Dev Pr(>Chi)
## NULL 1175 1040.18
## BusinessTravel 2 18.868 1173 1021.32 7.995e-05 ***
## Department 2 8.692 1171 1012.62 0.0129553 *
## EducationField 5 6.340 1166 1006.28 0.2744831
## Gender 1 0.637 1165 1005.65 0.4246324
## JobRole 8 69.556 1157 936.09 6.024e-12 ***
## OverTime 1 61.827 1156 874.26 3.751e-15 ***
## MaritalStatus 2 32.054 1154 842.21 1.095e-07 ***
## Age 1 5.387 1153 836.82 0.0202852 *
## DailyRate 1 3.178 1152 833.64 0.0746336 .
## DistanceFromHome 1 10.304 1151 823.34 0.0013276 **
## Education 4 1.753 1147 821.59 0.7810091
## EmployeeNumber 1 0.109 1146 821.48 0.7416105
## EnvironmentSatisfaction 3 25.105 1143 796.37 1.468e-05 ***
## HourlyRate 1 0.090 1142 796.28 0.7642039
## JobInvolvement 3 19.187 1139 777.10 0.0002501 ***
## JobLevel 4 40.262 1135 736.83 3.821e-08 ***
## JobSatisfaction 3 18.479 1132 718.35 0.0003503 ***
## MonthlyIncome 1 4.053 1131 714.30 0.0440873 *
## MonthlyRate 1 3.116 1130 711.18 0.0775034 .
## NumCompaniesWorked 1 15.350 1129 695.83 8.932e-05 ***
## PercentSalaryHike 1 1.739 1128 694.10 0.1873082
## PerformanceRating 1 0.004 1127 694.09 0.9465417
## RelationshipSatisfaction 3 10.130 1124 683.96 0.0174921 *
## StockOptionLevel 3 8.053 1121 675.91 0.0449197 *
## TotalWorkingYears 1 1.976 1120 673.93 0.1598558
## TrainingTimesLastYear 1 2.900 1119 671.03 0.0885629 .
## WorkLifeBalance 3 15.866 1116 655.17 0.0012078 **
## YearsAtCompany 1 0.917 1115 654.25 0.3382468
## YearsInCurrentRole 1 8.610 1114 645.64 0.0033438 **
## YearsSinceLastPromotion 1 7.425 1113 638.21 0.0064314 **
## YearsWithCurrManager 1 5.180 1112 633.03 0.0228466 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1If we take a look at the Residual Deviance column of ANOVA above, we can get a feel for how much impact a single variable has in the model. The ANOVA calculates the Residual Deviance (lack of fit) that each variable contributes to. Each variable is introduced in order of appearance in dataset (column-wise). For instance, starting from Null model (1040.18), when the first variable, “BusinessTravel” is introduced, there is a drop of Deviance of 18 points. Later variables may contribute more or less to the overall Deviance. The variables that drop the Deviance to the greatest degree, are the ones that may be heavily contributing to Attrition.
4. Fitting the model using STEP
We will use STEP as selection procedure (direction=“both”), which starts with a NULL model and adding one variable at the time, validating the impact. If the variable is not meaningful it will be discarded. This process is repeated until we go thru all the variables. (Long output ommited, only last few lines shown here for brevity).
## step(null.model.train, scope=list(upper=full.model.train), direction="both", test="Chisq", data=data.train)##
## . . .
## < many steps later >
## . . .
## Df Deviance AIC LRT Pr(>Chi)
## <none> 653.05 745.05
## ...
## - DistanceFromHome 1 668.28 758.28 15.224 9.547e-05 ***
## - JobSatisfaction 3 672.87 758.87 19.820 0.0001850 ***
## - JobInvolvement 3 673.25 759.25 20.195 0.0001546 ***
## - NumCompaniesWorked 1 672.21 762.21 19.153 1.207e-05 ***
## - BusinessTravel 2 675.76 763.76 22.705 1.174e-05 ***
## - EnvironmentSatisfaction 3 683.97 769.97 30.921 8.831e-07 ***
## - JobLevel 4 686.83 770.83 33.779 8.271e-07 ***
## - JobRole 8 704.21 780.21 51.153 2.452e-08 ***
## - OverTime 1 743.18 833.18 90.130 < 2.2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## . . .
## Degrees of Freedom: 1175 Total (i.e. Null); 1130 Residual
## Null Deviance: 1040
## Residual Deviance: 649.2 AIC: 741.2The output of the Step procedure above, shows most significant variables (in descending order). We will choose these top 9 (lowest p-value) variables for our Final Training Model.
5. Create Final Training Model with top variables (lowest p-value).
final.model.train<-glm(formula=Attrition~OverTime+JobRole+JobLevel+EnvironmentSatisfaction+BusinessTravel+NumCompaniesWorked+JobInvolvement+JobSatisfaction+DistanceFromHome, family = binomial(link="logit"), data=data.train)
summary(final.model.train)##
## Call:
## glm(formula = Attrition ~ OverTime + JobRole + JobLevel + EnvironmentSatisfaction +
## BusinessTravel + NumCompaniesWorked + JobInvolvement + JobSatisfaction +
## DistanceFromHome, family = binomial(link = "logit"), data = data.train)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.7672 -0.5318 -0.2972 -0.1384 2.9413
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.376960 0.809049 -0.466 0.641266
## OverTimeYes 1.703728 0.199136 8.556 < 2e-16 ***
## JobRoleHuman Resources 0.282504 0.679131 0.416 0.677426
## JobRoleLaboratory Technician 0.439641 0.592724 0.742 0.458250
## JobRoleManager -0.971051 0.989569 -0.981 0.326452
## JobRoleManufacturing Director 0.004675 0.569080 0.008 0.993446
## JobRoleResearch Director -2.384876 1.190992 -2.002 0.045239 *
## JobRoleResearch Scientist -0.545522 0.612119 -0.891 0.372821
## JobRoleSales Executive 1.296621 0.457348 2.835 0.004581 **
## JobRoleSales Representative 1.080430 0.647221 1.669 0.095051 .
## JobLevel2 -2.075147 0.420737 -4.932 8.13e-07 ***
## JobLevel3 -1.142405 0.497756 -2.295 0.021727 *
## JobLevel4 -1.998374 0.727256 -2.748 0.005999 **
## JobLevel5 -0.617082 1.160975 -0.532 0.595058
## EnvironmentSatisfaction2 -1.236091 0.298756 -4.137 3.51e-05 ***
## EnvironmentSatisfaction3 -1.098946 0.253043 -4.343 1.41e-05 ***
## EnvironmentSatisfaction4 -1.200238 0.258863 -4.637 3.54e-06 ***
## BusinessTravelTravel_Frequently 1.379870 0.428882 3.217 0.001294 **
## BusinessTravelTravel_Rarely 0.593573 0.401833 1.477 0.139632
## NumCompaniesWorked 0.107992 0.037124 2.909 0.003627 **
## JobInvolvement2 -1.063282 0.375174 -2.834 0.004595 **
## JobInvolvement3 -1.540777 0.354925 -4.341 1.42e-05 ***
## JobInvolvement4 -2.007050 0.475901 -4.217 2.47e-05 ***
## JobSatisfaction2 -0.263847 0.285172 -0.925 0.354851
## JobSatisfaction3 -0.407503 0.254084 -1.604 0.108755
## JobSatisfaction4 -1.076902 0.272780 -3.948 7.88e-05 ***
## DistanceFromHome 0.037731 0.011219 3.363 0.000771 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 1040.18 on 1175 degrees of freedom
## Residual deviance: 757.71 on 1149 degrees of freedom
## AIC: 811.71
##
## Number of Fisher Scoring iterations: 7Sort the model by p-value
#sort summary by p-value descending
tempDF<-as.data.frame(summary(final.model.train)$coefficients)
tempDF[order(tempDF$`Pr(>|z|)`),]## Estimate Std. Error z value Pr(>|z|)
## OverTimeYes 1.703728051 0.19913586 8.555606350 1.172508e-17
## JobLevel2 -2.075147485 0.42073695 -4.932173176 8.131977e-07
## EnvironmentSatisfaction4 -1.200237626 0.25886288 -4.636576725 3.542267e-06
## EnvironmentSatisfaction3 -1.098945522 0.25304310 -4.342918298 1.406024e-05
## JobInvolvement3 -1.540776733 0.35492533 -4.341129219 1.417523e-05
## JobInvolvement4 -2.007049927 0.47590144 -4.217364685 2.471741e-05
## EnvironmentSatisfaction2 -1.236091203 0.29875636 -4.137455719 3.511782e-05
## JobSatisfaction4 -1.076902408 0.27277952 -3.947885866 7.884437e-05
## DistanceFromHome 0.037731054 0.01121887 3.363179102 7.705036e-04
## BusinessTravelTravel_Frequently 1.379869980 0.42888209 3.217364400 1.293742e-03
## NumCompaniesWorked 0.107991628 0.03712439 2.908912829 3.626880e-03
## JobRoleSales Executive 1.296621357 0.45734763 2.835089280 4.581288e-03
## JobInvolvement2 -1.063282476 0.37517447 -2.834101357 4.595475e-03
## JobLevel4 -1.998374192 0.72725593 -2.747827989 5.999147e-03
## JobLevel3 -1.142405118 0.49775586 -2.295111348 2.172674e-02
## JobRoleResearch Director -2.384876479 1.19099203 -2.002428577 4.523866e-02
## JobRoleSales Representative 1.080429892 0.64722135 1.669335991 9.505081e-02
## JobSatisfaction3 -0.407503426 0.25408361 -1.603816278 1.087546e-01
## BusinessTravelTravel_Rarely 0.593573311 0.40183349 1.477162370 1.396321e-01
## JobRoleManager -0.971051006 0.98956937 -0.981286438 3.264515e-01
## JobSatisfaction2 -0.263847412 0.28517207 -0.925221783 3.548506e-01
## JobRoleResearch Scientist -0.545521561 0.61211892 -0.891201924 3.728209e-01
## JobRoleLaboratory Technician 0.439641378 0.59272358 0.741730868 4.582504e-01
## JobLevel5 -0.617081935 1.16097506 -0.531520405 5.950582e-01
## (Intercept) -0.376959731 0.80904868 -0.465929603 6.412659e-01
## JobRoleHuman Resources 0.282503674 0.67913145 0.415977897 6.774262e-01
## JobRoleManufacturing Director 0.004674672 0.56908005 0.008214437 9.934459e-016. Test Final Training Model Against Test data
We are looking for a measure of how accurate this Training Model is against a subset of the data.
predict.results <- predict(final.model.train,newdata=subset(data.test,select=c(2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32)),type='response')
predict.results <- ifelse(predict.results > 0.5,1,0)
misClasificError <- mean(predict.results)
print(paste('Accuracy',1-misClasificError))## [1] "Accuracy 0.918367346938776"We see a 91% accuracy of the model against test data, so we are good.
6. Run Final model against all data.
Now we ned to run the Final model against all the data.
final.model.alldata<-glm(formula=Attrition~OverTime+JobRole+JobLevel+EnvironmentSatisfaction+BusinessTravel+NumCompaniesWorked+JobInvolvement+JobSatisfaction+DistanceFromHome, family = binomial(link="logit"), data=attrition)
summary(final.model.alldata)##
## Call:
## glm(formula = Attrition ~ OverTime + JobRole + JobLevel + EnvironmentSatisfaction +
## BusinessTravel + NumCompaniesWorked + JobInvolvement + JobSatisfaction +
## DistanceFromHome, family = binomial(link = "logit"), data = attrition)
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -1.7212 -0.5375 -0.3002 -0.1340 3.0712
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -0.83400 0.71661 -1.164 0.244499
## OverTimeYes 1.76957 0.17760 9.964 < 2e-16 ***
## JobRoleHuman Resources 0.47126 0.61437 0.767 0.443041
## JobRoleLaboratory Technician 0.31962 0.53750 0.595 0.552085
## JobRoleManager -0.99151 0.92744 -1.069 0.285033
## JobRoleManufacturing Director 0.05172 0.51121 0.101 0.919414
## JobRoleResearch Director -2.22301 1.03063 -2.157 0.031010 *
## JobRoleResearch Scientist -0.44267 0.55256 -0.801 0.423059
## JobRoleSales Executive 1.34413 0.41115 3.269 0.001078 **
## JobRoleSales Representative 1.09145 0.58897 1.853 0.063863 .
## JobLevel2 -2.08559 0.38254 -5.452 4.98e-08 ***
## JobLevel3 -1.38643 0.45075 -3.076 0.002099 **
## JobLevel4 -2.16495 0.68233 -3.173 0.001509 **
## JobLevel5 -0.25018 1.02437 -0.244 0.807054
## EnvironmentSatisfaction2 -0.99411 0.25952 -3.831 0.000128 ***
## EnvironmentSatisfaction3 -1.09562 0.23010 -4.762 1.92e-06 ***
## EnvironmentSatisfaction4 -1.20054 0.23452 -5.119 3.07e-07 ***
## BusinessTravelTravel_Frequently 1.66670 0.38353 4.346 1.39e-05 ***
## BusinessTravelTravel_Rarely 0.85413 0.35785 2.387 0.016993 *
## NumCompaniesWorked 0.12553 0.03292 3.813 0.000137 ***
## JobInvolvement2 -0.97597 0.32700 -2.985 0.002839 **
## JobInvolvement3 -1.40635 0.30906 -4.550 5.35e-06 ***
## JobInvolvement4 -2.10620 0.43373 -4.856 1.20e-06 ***
## JobSatisfaction2 -0.42584 0.25427 -1.675 0.093976 .
## JobSatisfaction3 -0.48860 0.22383 -2.183 0.029042 *
## JobSatisfaction4 -1.08938 0.23978 -4.543 5.54e-06 ***
## DistanceFromHome 0.04245 0.01011 4.200 2.67e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 1298.58 on 1469 degrees of freedom
## Residual deviance: 946.73 on 1443 degrees of freedom
## AIC: 1000.7
##
## Number of Fisher Scoring iterations: 67. Obtain Odd Ratio of covariences
The Estimate in the final model is given in terms of Log Odds. Need to invert to obtain actual Odd Ratios of the top variables and sort output
Looking at the odd ratio table below, we can see that the Sales Representatives category has the highest odds of causing Attrition, this is, Sales representatives are 8 times more likely to have Attrition than any other category. The second highest covariate is Working Overtime. people who work overtime are 5 times more likely to have Atrition. People who travel frequently are 4x more likely to leave. Also Job Roles like Lab technician and HR are more likely to leave, based on the data.
| Covarience/Value | Odds Ratio |
|---|---|
| JobRoleSales Representative | 8.3669402 |
| OverTimeYes | 5.3426139 |
| BusinessTravelTravel_Frequently | 4.7865786 |
| JobRoleLaboratory Technician | 3.6904708 |
| JobRoleHuman Resources | 3.4551899 |
| JobRoleSales Executive | 3.1500722 |
| BusinessTravelTravel_Rarely | 2.2161456 |
| JobRoleResearch Scientist | 1.7691884 |
| JobRoleManager | 1.5547907 |
| NumCompaniesWorked | 1.119052 |
| DistanceFromHome | 1.0362319 |
Job Roles
Job Roles
Note: All analysis has been performed on employees only (No attrition participants)
Key Findings
1) Manager Job Role:
a) Only job role that exists in all departments. All other job roles are unique to a department.
b) Have the longest average duration at the company
2) Managers and Directors
a) Tend to be older than people in other job roles
b) Have higher Job Levels
c) Have higher Monthly Income
3) Manufacturing Director
a) The only Job Role where Males do not outnumber Females
4) Human Resource Manager
a) is the only Role which does not have ‘Low’ level of job involvement
b) No PhDs (characteristic in common with Sale Reps)
c) All HR Managers have some College level education (a characteristic unique to this Job Role)
5) Sales Reps and Reseach Scientists
a) have the lowest Job Levels
6) Sales Reps
a) is the only Job Role which has no employees with NumWorked = 9!!
b) have the lowest average duration at the company
c) change managers most frequently, while directors and managers tend to stay with the same manager for a longer time
d) No PhDs (characteristic in common with HR Managers)
e) the highest percentage of employees with college-level education (25%)
Departments
ggplot(data=as.data.frame(table(noattrit$Department,noattrit$JobRole,dnn=list("Department","JobRole"))), aes(x= reorder(JobRole,Freq),y=Freq, fill = Department)) +
geom_bar(stat="identity") + coord_flip() + ggtitle("Number of Employees per Job Role by Department") + xlab("") + ylab("")+theme(plot.title=element_text(hjust=0.5))+scale_fill_manual(values=alpha(c("steelblue4", "thistle4", "paleturquoise4")))
Linear Regression Modeling (See Experiment Tab) suggests Education Field, Age, Job Involvement, Gender, Job Level, Monthly Income, Number of Companies Worked, Years At Company and Years with Current Manager are different between of Job Roles
Education
ggplot(data=as.data.frame(table(noattrit$JobRole,noattrit$EducationField,dnn=list("JobRole","EducationField"))), aes(x= reorder(JobRole,Freq),y=Freq, fill = EducationField)) +
geom_bar(stat="identity",position="fill") + coord_flip() + ggtitle("Percent of Employees per Job Role by Education Field") + xlab("") + ylab("Percent of Employees")+theme(plot.title=element_text(hjust=0.5))+scale_fill_manual(values=alpha(c("steelblue4", "thistle4", "paleturquoise4", "midnightblue", "mistyrose4", "lightcyan3")))
Though not identified as a first order differentiator by linear regression, Education is an interesting parameter in regards to Job Role
ggplot(data=as.data.frame(table(noattrit$JobRole,noattrit$Education,dnn=list("JobRole","Education"))), aes(x= reorder(JobRole,Freq),y=Freq, fill = Education)) +
geom_bar(stat="identity",position="fill") + coord_flip() + ggtitle("Education by Job Role \nNo PhD Sales Reps nor HR Managers \nAll HR Managers Have Some College") + xlab("") + ylab("Percent of Employees")+theme(plot.title=element_text(hjust=0.5))+scale_fill_manual(values=alpha(c("steelblue4", "thistle4", "paleturquoise4", "mistyrose4", "lightcyan3")))
Observations
All Departments have a mix of educational fields
People with Education in Marketing only work in the Sales Department
People with Education in Human Resources work in the Human Resources Department
R&D is predominantly composed of people with Life Sciences and Medical Education
No PhD Sales Reps nor HR Managers
All HR Managers Have Some College (unique from all other job roles)
Sales Reps have the highest percentage of employees with college-level education (25%)
Age
noattrit$JobRole <-with(noattrit, ifelse(noattrit$JobRole=="Manager", paste(noattrit$Department,noattrit$JobRole), as.character(noattrit$JobRole)))
ggplot(data=noattrit,aes(x= JobRole,y=Age)) +
geom_boxplot() + coord_flip() + ggtitle("Age by Job Role") + xlab("")+ylab("Age")+theme(plot.title=element_text(hjust=0.5))
#Descripitve Statistics of Employee Ages
summary(noattrit$Age)## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 18.00 31.00 36.00 37.56 43.00 60.00 results <- lm(noattrit$JRCode~noattrit$Age)
summary(results)##
## Call:
## lm(formula = noattrit$JRCode ~ noattrit$Age)
##
## Residuals:
## Min 1Q Median 3Q Max
## -4.0652 -2.0573 0.9379 1.9400 4.9427
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 5.0494530 0.2910012 17.352 <2e-16 ***
## noattrit$Age 0.0002632 0.0075984 0.035 0.972
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 2.31 on 1196 degrees of freedom
## (35 observations deleted due to missingness)
## Multiple R-squared: 1.003e-06, Adjusted R-squared: -0.0008351
## F-statistic: 0.0012 on 1 and 1196 DF, p-value: 0.9724Observations
1. Managers and Directors tend be older than people in other roles
2. The oldest employee is a Sales Executive (60 years old)
3. The youngest employee is a Laboratory Technician (18 years old)
Gender
ggplot(data=as.data.frame(table(noattrit$JobRole,noattrit$Gender,dnn=list("JobRole","Gender"))), aes(x= reorder(JobRole,Freq),y=Freq, fill = Gender)) +
geom_bar(stat="identity",position="fill") + coord_flip() + ggtitle("Males OutNumber Females in Most Job Roles \n (Delta is not statiscally significant)") + xlab("") + ylab("Percent of Employees")+theme(plot.title=element_text(hjust=0.5))+scale_fill_manual(values=alpha(c("steelblue4", "thistle4"))) + geom_hline(data=noattrit, aes(yintercept = .50), colour="black") 
Compensation
ggplot(data=noattrit,aes(x= JobRole,y=MonthlyIncome,group=JobRole)) + geom_boxplot() + coord_flip() + ggtitle("Managers and Directors Have Higher Compensation \n (p-value < .00001)") + xlab("")+ylab("Monthly Income")+theme(plot.title=element_text(hjust=0.5))
Job Levels
ggplot(data=as.data.frame(table(noattrit$JobRole,noattrit$JobLevel,dnn=list("JobRole","JobLevel"))), aes(x= reorder(JobRole,Freq),y=Freq, fill = JobLevel)) + geom_bar(stat="identity",position="fill") + coord_flip() + ggtitle("Managers & Directors Tend to Have Higher Job Levels") + xlab("") + ylab("Percent of Employees")+theme(plot.title=element_text(hjust=0.5))+scale_fill_manual(values=alpha(c("steelblue4", "thistle4", "paleturquoise4", "mistyrose4", "lightcyan3")))
results <- lm(noattrit$JRCode~noattrit$JobLevel)
summary(results)##
## Call:
## lm(formula = noattrit$JRCode ~ noattrit$JobLevel)
##
## Residuals:
## Min 1Q Median 3Q Max
## -4.526 -1.795 0.718 2.205 4.718
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 4.55089 0.14517 31.349 < 2e-16 ***
## noattrit$JobLevel 0.24371 0.06191 3.937 8.74e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 2.295 on 1196 degrees of freedom
## (35 observations deleted due to missingness)
## Multiple R-squared: 0.01279, Adjusted R-squared: 0.01197
## F-statistic: 15.5 on 1 and 1196 DF, p-value: 8.742e-05Observations:
1. Managers and Directors have higher Job Levels.
2. Sales Reps and Research Scientists have the lowest Job Levels.
Job Involvement
ggplot(data=as.data.frame(table(attrit$JobRole,attrit$JobInvolvement,dnn=list("JobRole","JobInvolvement"))), aes(x= reorder(JobRole,Freq),y=Freq, fill = JobInvolvement)) +
geom_bar(stat="identity",position="fill") + coord_flip() + ggtitle(" HR Managers are the only Role \nwithout Low Job Involvement") + xlab("") + ylab("Percent of Employees") +theme(plot.title=element_text(hjust=0.5))+scale_fill_manual(values=alpha(c("steelblue4", "thistle4", "paleturquoise4", "mistyrose4")))
Work Experience
ggplot(data=noattrit,aes(x= JobRole,y=noattrit$NumCompaniesWorked)) +
geom_boxplot() + coord_flip() + ggtitle("Many Employees have Experience at Several Companies") + xlab("")+ylab("Number of Companies") +theme(plot.title=element_text(hjust=0.5))
Observations:
Sales Rep is the only Job Role which has no employees with NumWorked = 9!!
Years at Company
ggplot(data=noattrit,aes(x= JobRole,y=noattrit$YearsAtCompany)) + geom_boxplot() + coord_flip() + ggtitle("Significant Differences between Tenure by Job Role \n(p-value=.0008)") + xlab("")+ylab("Number of Years at Company") +theme(plot.title=element_text(hjust=0.5))
results <- lm(noattrit$JRCode~noattrit$YearsAtCompany)
summary(results)##
## Call:
## lm(formula = noattrit$JRCode ~ noattrit$YearsAtCompany)
##
## Residuals:
## Min 1Q Median 3Q Max
## -4.710 -1.954 0.756 1.945 5.121
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 4.87899 0.10601 46.025 <2e-16 ***
## noattrit$YearsAtCompany 0.02517 0.01152 2.186 0.029 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 2.305 on 1196 degrees of freedom
## (35 observations deleted due to missingness)
## Multiple R-squared: 0.003979, Adjusted R-squared: 0.003146
## F-statistic: 4.778 on 1 and 1196 DF, p-value: 0.02902Observations
The Job Role “Sales Reps” has the lowest average duration at the company.
The Job Roles of Managers have the longest average duration at the company.
ggplot(data=noattrit,aes(x= JobRole,y=YearsWithCurrManager)) + geom_boxplot() + coord_flip() + ggtitle("Years With Current Manager Vary By Job Role \n (p-value = .0425)") + xlab("")+ylab("Number of Years")+theme(plot.title=element_text(hjust=0.5))
Observations:
Sales Reps change managers most frequently, while directors and managers tend to stay with the same manager for a longer time.
Company Specifics
Company Specifics
Key Findings
67% of the Employees in the Company in R&D
The average age of employees in 37.56 years, with the employees in HR average age being slightly higher
There are 46% more Male employees than Female employees. HR has the largest ratio of Males to Female at 2.6:1
Job Level is strongly correlated to Montly Income. High Job Level = Higher Monthly Incomes
Human Resource Managers are the only Role which does not have ‘Low’ level of job involvement
50% of Employees have been with the company <= 5 years
People with Marketing education only work in Sales, and people with HR education only work in HR. All departments have a mix of Educational Backgrounds
Everyone receives a Performance Rating of Excellent or Outstanding
Work Experience correlates to Job Level and therefore Monthly Income
Monthly Income is based on Work Experience, not Performance. (Job Level is correlated to Total Time Worked and Time At Company. Monthly Income is correlated to Job Level and Total Time Worked. All Performance Ratings are Excellent and Outstanding.)
The longer you work for a manager the longer the time between promotions
Departments
DepartmentDF <- as.data.frame(table(noattrit$Department))
colnames(DepartmentDF) <- c("Department","DepartmentCount")
DepartmentDF$CoDepartmntPercentage <-paste(100*as.numeric(format(DepartmentDF$DepartmentCount/colSums(DepartmentDF[2]),digits=1)),"%")
DepartmentDF$CoDepartmntPercentage <-paste(100*as.numeric(format(DepartmentDF$DepartmentCount/colSums(DepartmentDF[2]),digits=1)),"%")
print.data.frame(DepartmentDF[with(DepartmentDF, order(-DepartmentCount)),],row.names=FALSE)## Department DepartmentCount CoDepartmntPercentage
## Research & Development 828 67 %
## Sales 354 29 %
## Human Resources 51 4 %ggplot(data=as.data.frame(table(noattrit$Department,dnn=list("Department"))), aes(x= reorder(Department,Freq),y=Freq)) +
geom_bar(stat="identity", fill="steelblue4") + coord_flip() + ggtitle("Department Sizes \n(no attrition)") + xlab("") + ylab("Number of Employees")+theme(plot.title=element_text(hjust=0.5))
Age Distribution
options(width = 150)
ggplot(data=noattrit, aes(x=noattrit$Age,fill=noattrit$Department)) + geom_histogram()+theme(plot.title=element_text(hjust=0.5))+xlab("Age")+ylab("Number of Employees")+ggtitle("Age Distribution by Department")+scale_fill_manual(values=alpha(c("steelblue4", "thistle4", "paleturquoise4")))## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
cat("Summary of Age Distribution for the Company")## Summary of Age Distribution for the Companysummary(noattrit$Age)## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 18.00 31.00 36.00 37.56 43.00 60.00cat("Summary of Age Distributions by Department")## Summary of Age Distributions by Departmentaggdata <-aggregate(noattrit$Age, by=list(noattrit$Department),
FUN=summary, na.rm=TRUE)
print(aggdata)## Group.1 x.Min. x.1st Qu. x.Median x.Mean x.3rd Qu. x.Max.
## 1 Human Resources 24.00 33.50 39.00 39.63 44.50 59.00
## 2 Research & Development 18.00 31.00 36.00 37.62 43.25 60.00
## 3 Sales 18.00 31.00 36.00 37.14 42.75 60.00Gender Distribution by Department
ggplot(data=as.data.frame(table(noattrit$Department,noattrit$Gender,dnn=list("Department","Gender"))), aes(x= reorder(Department,Freq),y=Freq, fill = Gender)) + geom_bar(stat="identity",position="fill") + coord_flip() + ggtitle("Gender by Department") + xlab("") + ylab("Percent of Employees") + geom_hline(data=noattrit, aes(yintercept = .50), colour="black") +theme(plot.title=element_text(hjust=0.5))+scale_fill_manual(values=alpha(c("steelblue4", "thistle4")))
ggplot(data=noattrit, aes(x=noattrit$Gender,fill=Gender)) + geom_bar() + ggtitle("There are 46% more Males than Females in the Company \n(p-value=.04)") + xlab("Gender")+ylab("Number of Employees")+theme(plot.title=element_text(hjust=0.5))+scale_fill_manual(values=alpha(c("steelblue4", "thistle4")))
#Recode Department to a numeric for t-test analysis
noattrit$DeptCode[noattrit$Department=="Research & Development"] <- 1L
noattrit$DeptCode[noattrit$Department=="Sales"] <- 2L
noattrit$DeptCode[noattrit$Department=="Human Resources"] <- 3L
t.test(table(noattrit$DeptCode, noattrit$Gender))##
## One Sample t-test
##
## data: table(noattrit$DeptCode, noattrit$Gender)
## t = 2.7531, df = 5, p-value = 0.04016
## alternative hypothesis: true mean is not equal to 0
## 95 percent confidence interval:
## 13.62051 397.37949
## sample estimates:
## mean of x
## 205.5ggplot(data=noattrit,aes(x=noattrit$JobLevel, y=noattrit$MonthlyIncome)) + geom_point() + ggtitle("Monthly Income is correlated to Job Level \n(p-value=<.0001)") + ylab("Monthly Income") +xlab("Job Level") + geom_smooth(method = 'lm', se = FALSE, color="midnightblue")+theme(plot.title=element_text(hjust=0.5))+scale_fill_manual(values=alpha(c("steelblue4")))
results <- lm(noattrit$MonthlyIncome~noattrit$JobLevel)
summary(results)##
## Call:
## lm(formula = noattrit$MonthlyIncome ~ noattrit$JobLevel)
##
## Residuals:
## Min 1Q Median 3Q Max
## -5119.4 -1021.7 136.6 810.0 3763.0
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -1953.62 92.83 -21.05 <2e-16 ***
## noattrit$JobLevel 4094.33 38.37 106.72 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1505 on 1231 degrees of freedom
## Multiple R-squared: 0.9025, Adjusted R-squared: 0.9024
## F-statistic: 1.139e+04 on 1 and 1231 DF, p-value: < 2.2e-16Job Involvement
ggplot(data=as.data.frame(table(noattrit$Department,noattrit$JobInvolvement,dnn=list("Department","JobInvolvement"))), aes(x= reorder(Department,Freq),y=Freq, fill = JobInvolvement)) + geom_bar(stat="identity",position="fill") + coord_flip() + ggtitle("4.4% of the Company has a Low Level of Job Involvement") + xlab("") + ylab("") + geom_hline(data=noattrit, aes(yintercept = .9554), colour="black")+theme(plot.title=element_text(hjust=0.5))+scale_fill_manual(values=alpha(c("steelblue4", "thistle4", "paleturquoise4", "mistyrose4")))
t.test(table(noattrit$DeptCode, noattrit$JobInvolvement))##
## One Sample t-test
##
## data: table(noattrit$DeptCode, noattrit$JobInvolvement)
## t = 2.5003, df = 11, p-value = 0.02949
## alternative hypothesis: true mean is not equal to 0
## 95 percent confidence interval:
## 12.29995 193.20005
## sample estimates:
## mean of x
## 102.75Number of Years with Current Manager
ggplot(data=noattrit, aes(x=factor(noattrit$YearsWithCurrManager))) + geom_histogram(stat="count",position = "dodge", fill="steelblue4") +xlab("Years")+ylab("Number of Employees") + scale_x_discrete(name='Years With Current Manager') + ggtitle("Years With Current Manager")+theme(plot.title=element_text(hjust=0.5))## Warning: Ignoring unknown parameters: binwidth, bins, pad
Observations:
Multiple Modal Distribution
Education
cat("Percentage of Employees with No College-Level Education:",paste(100*as.numeric(format(nrow(as.data.frame(noattrit[( noattrit$Education == 1),]))/1233,digits=1)),"%"))## Percentage of Employees with No College-Level Education: 10 %ggplot(data=noattrit, aes(x=factor(noattrit$Education))) + geom_histogram(stat="count",position = "dodge", fill="steelblue4") +xlab(" ") + scale_x_discrete(name='Education') + ggtitle("Education Level")+theme(plot.title=element_text(hjust=0.5))## Warning: Ignoring unknown parameters: binwidth, bins, pad
Compensation is based on Work Experience not Performance
Job Level vs. Total Working Years
ggplot(data=noattrit,aes(x= JobLevel,y=noattrit$TotalWorkingYears, group=JobLevel)) + geom_boxplot() + coord_flip() + ggtitle("Work Experience Correlates to Job Level") + xlab("Job Level")+ylab("Years")+theme(plot.title=element_text(hjust=0.5))
Working Years vs. Monthly Income
ggplot(data=noattrit,aes(x=noattrit$TotalWorkingYears,
y=noattrit$MonthlyIncome,col=factor(noattrit$JobLevel))) +
geom_point() + ggtitle("Total Working Years is Correlated to Job Level \n(p-value <.00001)") + xlab("Working Years")+ylab("Monthly Income")+theme(plot.title=element_text(hjust=0.5))+scale_color_manual(values=alpha(c("steelblue4", "thistle4", "paleturquoise4", "mistyrose4", "lightcyan3")))
results <- lm(noattrit$JobLevel~noattrit$TotalWorkingYears)
summary(results)##
## Call:
## lm(formula = noattrit$JobLevel ~ noattrit$TotalWorkingYears)
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.06151 -0.48633 0.06362 0.40116 2.17613
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.811251 0.036340 22.32 <2e-16 ***
## noattrit$TotalWorkingYears 0.112513 0.002564 43.89 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.6984 on 1231 degrees of freedom
## Multiple R-squared: 0.6101, Adjusted R-squared: 0.6098
## F-statistic: 1926 on 1 and 1231 DF, p-value: < 2.2e-16Total Time at Company vs. Monthly Income
ggplot(data=noattrit,aes(x=noattrit$YearsAtCompany, y=noattrit$MonthlyIncome,col=factor(noattrit$JobLevel))) +
geom_point() + ggtitle("Years At Company vs Monthly Income by JobLevel") + xlab("Years at Company")+ylab("Monthly Income")+theme(plot.title=element_text(hjust=0.5))+scale_color_manual(values=alpha(c("steelblue4", "thistle4", "paleturquoise4", "mistyrose4", "lightcyan3")))
results <- lm(noattrit$MonthlyIncome ~ noattrit$JobLevel + noattrit$TotalWorkingYears + noattrit$YearsAtCompany)
summary(results)##
## Call:
## lm(formula = noattrit$MonthlyIncome ~ noattrit$JobLevel + noattrit$TotalWorkingYears +
## noattrit$YearsAtCompany)
##
## Residuals:
## Min 1Q Median 3Q Max
## -5295.9 -1004.0 90.5 849.6 3964.0
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -1943.842 92.416 -21.034 < 2e-16 ***
## noattrit$JobLevel 3848.637 60.956 63.138 < 2e-16 ***
## noattrit$TotalWorkingYears 53.394 9.388 5.688 1.61e-08 ***
## noattrit$YearsAtCompany -15.735 8.680 -1.813 0.0701 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1487 on 1229 degrees of freedom
## Multiple R-squared: 0.905, Adjusted R-squared: 0.9047
## F-statistic: 3901 on 3 and 1229 DF, p-value: < 2.2e-16Performance Rating
ggplot(data=noattrit, aes(x=factor(noattrit$PerformanceRating))) + geom_histogram(stat="count",position = "dodge", fill="steelblue4") +xlab(" ") + scale_x_discrete(name='Performance Rating') + ggtitle("Everyone Receives a Performance Rating of Excellent or Outstanding")+theme(plot.title=element_text(hjust=0.5))## Warning: Ignoring unknown parameters: binwidth, bins, pad
ggplot(data=noattrit,aes(x= PerformanceRating, y=noattrit$MonthlyIncome, group=PerformanceRating)) + geom_boxplot() + ggtitle("Monthly Income by Performance Rating") + xlab("Performance Rating")+ylab("Monthly Income")+theme(plot.title=element_text(hjust=0.5))
Years Since Promotion
ggplot(data=noattrit,aes(x=noattrit$YearsWithCurrManager, y=noattrit$YearsSinceLastPromotion)) +
geom_point() + ggtitle("The Longer You Stay with a Manager the Longer Time Between Promotions") + xlab("Years With Current Manager")+ylab("Years Since Last Promotion") + stat_smooth(method="lm", se=FALSE, color="midnightblue")+theme(plot.title=element_text(hjust=0.5))+scale_color_manual(values=alpha(c("steelblue4")))
results <- lm(noattrit$YearsSinceLastPromotion ~ noattrit$YearsWithCurrManager)
summary(results)##
## Call:
## lm(formula = noattrit$YearsSinceLastPromotion ~ noattrit$YearsWithCurrManager)
##
## Residuals:
## Min 1Q Median 3Q Max
## -6.8662 -1.6387 -0.3318 0.7969 14.6682
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.33183 0.12694 2.614 0.00906 **
## noattrit$YearsWithCurrManager 0.43563 0.02245 19.407 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 2.832 on 1231 degrees of freedom
## Multiple R-squared: 0.2343, Adjusted R-squared: 0.2337
## F-statistic: 376.6 on 1 and 1231 DF, p-value: < 2.2e-16ggplot(data=noattrit,aes(x= JobLevel,y=noattrit$YearsSinceLastPromotion, group=JobLevel)) + geom_boxplot() + coord_flip() + ggtitle("Time Since Last Promotion Increases as Job Level Increases") + xlab("Job Level")+ylab("Years")+theme(plot.title=element_text(hjust=0.5))
Percent Salary Hike
ggplot(data=noattrit,aes(x= JobLevel,y=noattrit$PercentSalaryHike, group=JobLevel)) + geom_boxplot() + coord_flip() + ggtitle("All Job Levels Received About 15% Average Salary Increase \nwith roughly similar distributions") + xlab("Job Level")+ylab("% Salary Increase")+theme(plot.title=element_text(hjust=0.5))
Job Satisfaction
Job Satisfaction
Key Findings
Job Satisfaction decreaseas as Hourly rate
PhDs have the highest satisfaction
People with Bachelor Degrees are the most dissatified
Satisfaction in consistent by Gender
Satisfaction varies by Job Role with R&D Manager having the lowest satisfaction and Sales Reps and HR Managers having the highest
Hourly Rate
Suggested by Linear Regression Modeling (See Experimentation Tab) as weak factor for Job Satisfaction (other potential factors were identified but did not pass on review)
ggplot(data=noattrit,aes(x= JobSatisfaction,y=noattrit$HourlyRate, group=JobSatisfaction)) + geom_boxplot() + coord_flip() + ggtitle("Job Satisfaction Decreases as Hourly Rate Decreases \np-value=.05") + xlab("Job Satisfaction")+ylab("Hourly Rate")+theme(plot.title=element_text(hjust=0.5))
Visual inspection of the Top 10 common demographics suggests exploring Department, Education Field, Gender and Job Role as factors related to Job Satisfaction
#recode JobSatisfaction for analysis
noattrit$JSCode[noattrit$JobSatisfaction==1] <- "Low"
noattrit$JSCode[noattrit$JobSatisfaction==2 |noattrit$JobSatisfaction==3 | noattrit$JobSatisfaction==4] <- "High"
noattrit$JSCodeNum[noattrit$JobSatisfaction==2 |noattrit$JobSatisfaction==3 | noattrit$JobSatisfaction==4] <- 2L
noattrit$JSCodeNum[noattrit$JobSatisfaction==1] <- 1LEducation
ggplot(data=as.data.frame(table(noattrit$Education,noattrit$JobSatisfaction,dnn=list("Education","JobSatisfaction"))), aes(x=Education,y=Freq, fill = JobSatisfaction)) + geom_bar(stat="identity",position="fill") + ggtitle("PhDs Have the Highest Job Satisfaction \n People with Bachelor Degrees Have the Lowest Job Satisfaction \n(p-value=.01263)") + xlab("Education Level") + ylab("Percent of Employees") + geom_abline(slope=0, intercept=0.5, col = "black",lty=2) + coord_flip()+theme(plot.title=element_text(hjust=0.5))+scale_fill_manual(values=alpha(c("steelblue4", "thistle4", "paleturquoise4", "mistyrose4")))
ggplot(data=as.data.frame(table(noattrit$Education,noattrit$JSCode,dnn=list("Education","JobSatisfaction"))), aes(x=Education,y=Freq, fill = JobSatisfaction)) + geom_bar(stat="identity",position="fill") + ggtitle("Job Satisfaction Varies by Education") + xlab("Education Level") + ylab("Percent of Employees") + coord_flip()+theme(plot.title=element_text(hjust=0.5))+scale_fill_manual(values=alpha(c("steelblue4", "thistle4")))
t.test(table(noattrit$JSCodeNum, noattrit$Education))##
## One Sample t-test
##
## data: table(noattrit$JSCodeNum, noattrit$Education)
## t = 3.1046, df = 9, p-value = 0.01263
## alternative hypothesis: true mean is not equal to 0
## 95 percent confidence interval:
## 33.45881 213.14119
## sample estimates:
## mean of x
## 123.3Gender
ggplot(data=as.data.frame(table(noattrit$Gender,noattrit$JobSatisfaction,dnn=list("Gender","JobSatisfaction"))), aes(x=Gender,y=Freq, fill = JobSatisfaction)) +
geom_bar(stat="identity",position="fill") + ggtitle("Job Satisfaction is Consistent Across Genders") + xlab("Gender") + ylab("Percent of Employees") + geom_abline(slope=0, intercept=0.5, col = "black",lty=2)+theme(plot.title=element_text(hjust=0.5))+scale_fill_manual(values=alpha(c("steelblue4", "thistle4", "paleturquoise4", "mistyrose4")))
Job Role
ggplot(data=as.data.frame(table(noattrit$JobRole,noattrit$JobSatisfaction,dnn=list("JobRole","JobSatisfaction"))), aes(x=JobRole,y=Freq, fill = JobSatisfaction)) + geom_bar(stat="identity",position="fill") + ggtitle("Job Satisfaction is Varies by Job Roles \nTop Dissatisfied Roles:R&D Director \nMost Satsifed: Sales Reps and HR Manager") + xlab("") + ylab("Percent of Employees") + geom_abline(slope=0, intercept=0.9, col = "black",lty=2) + coord_flip()+theme(plot.title=element_text(hjust=0.5))+scale_fill_manual(values=alpha(c("steelblue4", "thistle4", "paleturquoise4", "mistyrose4")))
t.test(table(noattrit$JSCodeNum, noattrit$JobRole))##
## One Sample t-test
##
## data: table(noattrit$JSCodeNum, noattrit$JobRole)
## t = 4.1555, df = 21, p-value = 0.000448
## alternative hypothesis: true mean is not equal to 0
## 95 percent confidence interval:
## 27.99778 84.09313
## sample estimates:
## mean of x
## 56.04545Interesting Findings
Interesting Findings
Key Findings
NumCompaniesWorked has evidence of ambiguity in the data and is perhaps miscoded or inaccurate. (See discussion below)
No relationship between the difference compensation rates. Would have expected a multiplicative relationship between DailyRate and Hourly Rate, or Daily Rate and Monthly Rate, etc.
Observatons
Strong Correlation between observation incoming order and employee number
Number of Companies Worked
summary(noattrit$NumCompaniesWorked)## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 0.000 1.000 2.000 2.646 4.000 9.000ggplot(data=noattrit, aes(x=factor(noattrit$NumCompaniesWorked))) + geom_histogram(stat="count",position = "dodge", fill="steelblue4") +xlab("Number CompaniesWorked") + scale_x_discrete(name='Number of Companies') +ggtitle("Number of Companies Worked")+ylab("Number of Employees")+theme(plot.title=element_text(hjust=0.5))## Warning: Ignoring unknown parameters: binwidth, bins, pad
Observations
There maybe confusion in the coding of this variable. What does 0 mean? Since these are employees at this company everyone has worked at least one company. Or maybe the variable is to be interpreted as the number of companies worked other than this one. Seems unlikely as the most frequent value are
Compensation Rates - No correlations
Monthly Income vs. Monthly Rate
ggplot(data=noattrit,aes(x=noattrit$MonthlyRate, y=noattrit$MonthlyIncome)) +
geom_point() + ggtitle(" No Correlation between Monthly Income vs Monthly Rate \n p-value=.261") + xlab("Monthly Rate")+ylab("Monthly Income") + stat_smooth(method="lm", se=FALSE, color="midnightblue")+theme(plot.title=element_text(hjust=0.5))
result <- lm(noattrit$MonthlyIncome ~ noattrit$MonthlyRate)
summary(result)##
## Call:
## lm(formula = noattrit$MonthlyIncome ~ noattrit$MonthlyRate)
##
## Residuals:
## Min 1Q Median 3Q Max
## -5979 -3597 -1645 1986 13353
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 6.522e+03 3.079e+02 21.180 <2e-16 ***
## noattrit$MonthlyRate 2.175e-02 1.933e-02 1.126 0.261
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 4818 on 1231 degrees of freedom
## Multiple R-squared: 0.001028, Adjusted R-squared: 0.0002168
## F-statistic: 1.267 on 1 and 1231 DF, p-value: 0.2605ggplot(data=noattrit,aes(x=noattrit$DailyRate, y=noattrit$MonthlyIncome)) +
geom_point() + ggtitle(" No Correlation between Monthly Income vs Daily Rate") + xlab("Daily Rate")+ylab("Monthly Income") + stat_smooth(method="lm", se=FALSE, color="midnightblue")+theme(plot.title=element_text(hjust=0.5))
ggplot(data=noattrit,aes(x=noattrit$DailyRate, y=noattrit$MonthlyRate)) +
geom_point() + ggtitle(" No Correlation between Monthly Rate & Daily Rate") + xlab("Daily Rate")+ylab("Monthly Rate") + stat_smooth(method="lm", se=FALSE, color="midnightblue")+theme(plot.title=element_text(hjust=0.5))
ggplot(data=noattrit,aes(x=noattrit$HourlyRate, y=noattrit$DailyRate)) +
geom_point() + ggtitle(" No Correlation between Hourly Rate & Daily Rate") + xlab("Hourly Rate")+ylab("Daily Rate") + stat_smooth(method="lm", se=FALSE, color="midnightblue")+theme(plot.title=element_text(hjust=0.5))
ggplot(data=noattrit,aes(x=noattrit$YearsAtCompany, y=noattrit$DailyRate,col=factor(noattrit$JobLevel))) +
geom_point() + ggtitle("Daily Rate vs Years At Company by JobLevel") + xlab("Years")+ylab("Daily Rate")+theme(plot.title=element_text(hjust=0.5))+scale_color_manual(values=alpha(c("steelblue4", "thistle4", "paleturquoise4", "mistyrose4", "lightcyan3")))
The strong correlations found between TimeAtCompany & Total Working Time with Monthly Income are not evident with DailyRate, Hourly Rate nor Monthly Rate.
Attempted to determine relationships using log-log, linear-log, log-linear transformations and multiplicative models.
Experimentation
Experimentation
1.Employee Number - a recoded start date?….No
2.Linear Regression Modeling to id first order variables to differentiate factor variables
Employee Number is strongly correlated to the index number of the incoming data. Perhaps the employee number is a time ordered variable related to employee start date?
ggplot(data=noattrit,aes(x=as.integer(rownames(noattrit)), y=noattrit$EmployeeNumber)) +
geom_point() + ggtitle("Index Number vs Employee Number") + ylab("Employee Number")+ xlab("Index Number")+theme(plot.title=element_text(hjust=0.5))
ggplot(data=noattrit,aes(x=noattrit$EmployeeNumber, y=noattrit$YearsAtCompany)) +
geom_point() + ggtitle("Years At Company vs Employee \nEmployee Number is not related to start date") + xlab("Employee Number")+ylab("Years")+theme(plot.title=element_text(hjust=0.5))
Used Linear Regression Modeling to determine variables of significance between Job Roles
Important first order terms in the model are: EducationField, Age, JobLevel, MonthlyIncome, Gender, JobInvolvement, NumCompaniesWorked, YearsAtCompany, YearsWithCurrManager
options(width = 150)
#Run a linear regression model using all variables to id the important first order terms
jr.lm <- lm(noattrit$JRCode ~ noattrit$BusinessTravel+ noattrit$EducationField+ noattrit$Gender+ noattrit$OverTime+ noattrit$MaritalStatus+ noattrit$Age+ noattrit$DailyRate+
noattrit$DistanceFromHome+ noattrit$Education+ noattrit$EmployeeNumber+noattrit$EnvironmentSatisfaction+ noattrit$HourlyRate+ noattrit$JobInvolvement+ noattrit$JobLevel+ noattrit$JobSatisfaction+
noattrit$MonthlyIncome+ noattrit$MonthlyRate+ noattrit$NumCompaniesWorked+ noattrit$PercentSalaryHike+ noattrit$PerformanceRating+ noattrit$RelationshipSatisfaction+ noattrit$StockOptionLevel+
noattrit$TotalWorkingYears+ noattrit$TrainingTimesLastYear+ noattrit$WorkLifeBalance+ noattrit$YearsAtCompany+ noattrit$YearsInCurrentRole+ noattrit$YearsSinceLastPromotion+ noattrit$YearsWithCurrManager)
summary(jr.lm)##
## Call:
## lm(formula = noattrit$JRCode ~ noattrit$BusinessTravel + noattrit$EducationField +
## noattrit$Gender + noattrit$OverTime + noattrit$MaritalStatus +
## noattrit$Age + noattrit$DailyRate + noattrit$DistanceFromHome +
## noattrit$Education + noattrit$EmployeeNumber + noattrit$EnvironmentSatisfaction +
## noattrit$HourlyRate + noattrit$JobInvolvement + noattrit$JobLevel +
## noattrit$JobSatisfaction + noattrit$MonthlyIncome + noattrit$MonthlyRate +
## noattrit$NumCompaniesWorked + noattrit$PercentSalaryHike +
## noattrit$PerformanceRating + noattrit$RelationshipSatisfaction +
## noattrit$StockOptionLevel + noattrit$TotalWorkingYears +
## noattrit$TrainingTimesLastYear + noattrit$WorkLifeBalance +
## noattrit$YearsAtCompany + noattrit$YearsInCurrentRole + noattrit$YearsSinceLastPromotion +
## noattrit$YearsWithCurrManager)
##
## Residuals:
## Min 1Q Median 3Q Max
## -5.5355 -1.6075 -0.1949 1.8112 4.7772
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 4.410e+00 1.087e+00 4.056 5.33e-05 ***
## noattrit$BusinessTravelTravel_Frequently -1.847e-01 2.466e-01 -0.749 0.45416
## noattrit$BusinessTravelTravel_Rarely -9.590e-02 2.067e-01 -0.464 0.64281
## noattrit$EducationFieldLife Sciences 1.170e+00 5.093e-01 2.297 0.02180 *
## noattrit$EducationFieldMarketing 2.600e+00 5.437e-01 4.782 1.95e-06 ***
## noattrit$EducationFieldMedical 1.083e+00 5.123e-01 2.114 0.03469 *
## noattrit$EducationFieldOther 1.002e+00 5.646e-01 1.774 0.07633 .
## noattrit$EducationFieldTechnical Degree 1.435e+00 5.486e-01 2.616 0.00901 **
## noattrit$GenderMale -2.214e-01 1.319e-01 -1.678 0.09361 .
## noattrit$OverTimeYes -4.345e-02 1.551e-01 -0.280 0.77937
## noattrit$MaritalStatusMarried 1.003e-01 1.698e-01 0.591 0.55479
## noattrit$MaritalStatusSingle 2.164e-01 2.398e-01 0.902 0.36701
## noattrit$Age -1.781e-02 1.024e-02 -1.739 0.08234 .
## noattrit$DailyRate -2.736e-05 1.609e-04 -0.170 0.86496
## noattrit$DistanceFromHome -4.919e-03 8.074e-03 -0.609 0.54250
## noattrit$Education -5.459e-03 6.475e-02 -0.084 0.93283
## noattrit$EmployeeNumber -1.061e-04 1.064e-04 -0.998 0.31872
## noattrit$EnvironmentSatisfaction -1.516e-02 6.083e-02 -0.249 0.80328
## noattrit$HourlyRate -2.095e-03 3.171e-03 -0.661 0.50902
## noattrit$JobInvolvement 2.160e-01 9.312e-02 2.320 0.02051 *
## noattrit$JobLevel -1.073e+00 1.932e-01 -5.555 3.44e-08 ***
## noattrit$JobSatisfaction 4.070e-02 5.940e-02 0.685 0.49335
## noattrit$MonthlyIncome 3.457e-04 4.427e-05 7.810 1.27e-14 ***
## noattrit$MonthlyRate 3.592e-06 9.061e-06 0.396 0.69182
## noattrit$NumCompaniesWorked -6.608e-02 2.965e-02 -2.229 0.02603 *
## noattrit$PercentSalaryHike -5.725e-03 2.766e-02 -0.207 0.83609
## noattrit$PerformanceRating 8.678e-02 2.784e-01 0.312 0.75531
## noattrit$RelationshipSatisfaction -3.219e-03 6.047e-02 -0.053 0.95755
## noattrit$StockOptionLevel 3.163e-02 1.037e-01 0.305 0.76049
## noattrit$TotalWorkingYears 1.107e-03 1.807e-02 0.061 0.95118
## noattrit$TrainingTimesLastYear -1.385e-02 5.001e-02 -0.277 0.78181
## noattrit$WorkLifeBalance -4.880e-02 9.561e-02 -0.510 0.60989
## noattrit$YearsAtCompany 4.173e-02 2.366e-02 1.764 0.07807 .
## noattrit$YearsInCurrentRole -2.674e-02 2.927e-02 -0.913 0.36118
## noattrit$YearsSinceLastPromotion 1.450e-02 2.607e-02 0.556 0.57803
## noattrit$YearsWithCurrManager -7.326e-02 2.980e-02 -2.459 0.01409 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 2.206 on 1162 degrees of freedom
## (35 observations deleted due to missingness)
## Multiple R-squared: 0.114, Adjusted R-squared: 0.08734
## F-statistic: 4.273 on 35 and 1162 DF, p-value: 6.053e-15#Check revised model fit
jr.lm <- lm(noattrit$JRCode ~noattrit$EducationField+ noattrit$Age + noattrit$JobLevel+ noattrit$MonthlyIncome+ noattrit$Gender+ noattrit$JobInvolvement+ noattrit$NumCompaniesWorked+ noattrit$YearsAtCompany+ noattrit$YearsWithCurrManager)
summary(jr.lm)##
## Call:
## lm(formula = noattrit$JRCode ~ noattrit$EducationField + noattrit$Age +
## noattrit$JobLevel + noattrit$MonthlyIncome + noattrit$Gender +
## noattrit$JobInvolvement + noattrit$NumCompaniesWorked + noattrit$YearsAtCompany +
## noattrit$YearsWithCurrManager)
##
## Residuals:
## Min 1Q Median 3Q Max
## -5.3740 -1.5687 -0.2193 1.8478 4.8356
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 4.276e+00 6.369e-01 6.714 2.93e-11 ***
## noattrit$EducationFieldLife Sciences 1.141e+00 5.016e-01 2.274 0.02313 *
## noattrit$EducationFieldMarketing 2.554e+00 5.357e-01 4.768 2.09e-06 ***
## noattrit$EducationFieldMedical 1.057e+00 5.037e-01 2.098 0.03610 *
## noattrit$EducationFieldOther 9.505e-01 5.568e-01 1.707 0.08809 .
## noattrit$EducationFieldTechnical Degree 1.389e+00 5.399e-01 2.573 0.01021 *
## noattrit$Age -1.773e-02 8.586e-03 -2.065 0.03915 *
## noattrit$JobLevel -1.073e+00 1.881e-01 -5.702 1.50e-08 ***
## noattrit$MonthlyIncome 3.465e-04 4.324e-05 8.015 2.63e-15 ***
## noattrit$GenderMale -2.246e-01 1.298e-01 -1.731 0.08378 .
## noattrit$JobInvolvement 2.057e-01 9.152e-02 2.247 0.02481 *
## noattrit$NumCompaniesWorked -6.591e-02 2.820e-02 -2.337 0.01958 *
## noattrit$YearsAtCompany 3.656e-02 1.897e-02 1.927 0.05418 .
## noattrit$YearsWithCurrManager -7.996e-02 2.815e-02 -2.840 0.00459 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 2.191 on 1184 degrees of freedom
## (35 observations deleted due to missingness)
## Multiple R-squared: 0.1094, Adjusted R-squared: 0.09957
## F-statistic: 11.18 on 13 and 1184 DF, p-value: < 2.2e-16The fit is not great, but perhaps sufficient enough to identify the important fit order terms to explore Job Roles
Linear Regression Modeling for Job Satisfaction first order factors
js.lm <- lm(noattrit$JobSatisfaction ~ noattrit$BusinessTravel+ noattrit$EducationField+ noattrit$Gender+ noattrit$OverTime+ noattrit$MaritalStatus+ noattrit$Age+ noattrit$DailyRate+ noattrit$DistanceFromHome+ noattrit$Education+ noattrit$EmployeeNumber+ noattrit$EnvironmentSatisfaction+ noattrit$HourlyRate+ noattrit$JobInvolvement+ noattrit$JobLevel+ noattrit$JobRole+ noattrit$MonthlyIncome+ noattrit$MonthlyRate+ noattrit$NumCompaniesWorked+ noattrit$PercentSalaryHike+ noattrit$PerformanceRating+ noattrit$RelationshipSatisfaction+ noattrit$StockOptionLevel+ noattrit$TotalWorkingYears+ noattrit$TrainingTimesLastYear+ noattrit$WorkLifeBalance+ noattrit$YearsAtCompany+ noattrit$YearsInCurrentRole+ noattrit$YearsSinceLastPromotion+ noattrit$YearsWithCurrManager)
summary(js.lm)##
## Call:
## lm(formula = noattrit$JobSatisfaction ~ noattrit$BusinessTravel +
## noattrit$EducationField + noattrit$Gender + noattrit$OverTime +
## noattrit$MaritalStatus + noattrit$Age + noattrit$DailyRate +
## noattrit$DistanceFromHome + noattrit$Education + noattrit$EmployeeNumber +
## noattrit$EnvironmentSatisfaction + noattrit$HourlyRate +
## noattrit$JobInvolvement + noattrit$JobLevel + noattrit$JobRole +
## noattrit$MonthlyIncome + noattrit$MonthlyRate + noattrit$NumCompaniesWorked +
## noattrit$PercentSalaryHike + noattrit$PerformanceRating +
## noattrit$RelationshipSatisfaction + noattrit$StockOptionLevel +
## noattrit$TotalWorkingYears + noattrit$TrainingTimesLastYear +
## noattrit$WorkLifeBalance + noattrit$YearsAtCompany + noattrit$YearsInCurrentRole +
## noattrit$YearsSinceLastPromotion + noattrit$YearsWithCurrManager)
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.3281 -0.8023 0.1658 1.0144 1.7461
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.388e+00 5.759e-01 5.884 5.2e-09 ***
## noattrit$BusinessTravelTravel_Frequently 3.423e-05 1.213e-01 0.000 0.9998
## noattrit$BusinessTravelTravel_Rarely -1.124e-01 1.015e-01 -1.107 0.2686
## noattrit$EducationFieldLife Sciences -9.068e-02 3.226e-01 -0.281 0.7787
## noattrit$EducationFieldMarketing -3.070e-01 3.443e-01 -0.892 0.3727
## noattrit$EducationFieldMedical -2.194e-01 3.244e-01 -0.676 0.4991
## noattrit$EducationFieldOther -2.565e-01 3.439e-01 -0.746 0.4559
## noattrit$EducationFieldTechnical Degree -1.939e-01 3.393e-01 -0.571 0.5678
## noattrit$GenderMale 8.196e-02 6.457e-02 1.269 0.2046
## noattrit$OverTimeYes 1.514e-01 7.572e-02 2.000 0.0457 *
## noattrit$MaritalStatusMarried 6.284e-02 8.295e-02 0.758 0.4488
## noattrit$MaritalStatusSingle 2.900e-01 1.170e-01 2.479 0.0133 *
## noattrit$Age 5.861e-04 5.039e-03 0.116 0.9074
## noattrit$DailyRate 7.588e-05 7.867e-05 0.965 0.3350
## noattrit$DistanceFromHome 2.969e-03 3.932e-03 0.755 0.4503
## noattrit$Education 6.693e-03 3.164e-02 0.211 0.8325
## noattrit$EmployeeNumber -6.788e-05 5.215e-05 -1.302 0.1933
## noattrit$EnvironmentSatisfaction -3.693e-02 2.976e-02 -1.241 0.2150
## noattrit$HourlyRate -3.215e-03 1.548e-03 -2.076 0.0381 *
## noattrit$JobInvolvement -5.204e-02 4.582e-02 -1.136 0.2563
## noattrit$JobLevel 2.744e-02 1.039e-01 0.264 0.7918
## noattrit$JobRoleHuman Resources -1.049e-01 2.361e-01 -0.444 0.6569
## noattrit$JobRoleHuman Resources Manager -7.513e-02 4.247e-01 -0.177 0.8596
## noattrit$JobRoleLaboratory Technician 1.770e-02 1.447e-01 0.122 0.9027
## noattrit$JobRoleManufacturing Director -9.222e-02 1.380e-01 -0.668 0.5041
## noattrit$JobRoleResearch & Development Manager -4.119e-02 2.440e-01 -0.169 0.8660
## noattrit$JobRoleResearch Director -7.286e-02 2.153e-01 -0.338 0.7351
## noattrit$JobRoleResearch Scientist 5.290e-02 1.407e-01 0.376 0.7070
## noattrit$JobRoleSales Executive 7.784e-02 1.301e-01 0.598 0.5497
## noattrit$JobRoleSales Manager 4.968e-02 2.636e-01 0.188 0.8506
## noattrit$JobRoleSales Representative 8.984e-02 2.031e-01 0.442 0.6583
## noattrit$MonthlyIncome 4.124e-06 2.776e-05 0.149 0.8819
## noattrit$MonthlyRate -5.035e-06 4.429e-06 -1.137 0.2558
## noattrit$NumCompaniesWorked -2.026e-02 1.441e-02 -1.406 0.1600
## noattrit$PercentSalaryHike 1.693e-02 1.361e-02 1.244 0.2137
## noattrit$PerformanceRating -3.150e-02 1.368e-01 -0.230 0.8180
## noattrit$RelationshipSatisfaction -4.988e-02 2.961e-02 -1.685 0.0923 .
## noattrit$StockOptionLevel 5.927e-02 5.093e-02 1.164 0.2447
## noattrit$TotalWorkingYears -3.949e-03 8.865e-03 -0.446 0.6560
## noattrit$TrainingTimesLastYear -5.648e-03 2.459e-02 -0.230 0.8184
## noattrit$WorkLifeBalance -2.412e-02 4.651e-02 -0.519 0.6041
## noattrit$YearsAtCompany -8.128e-05 1.095e-02 -0.007 0.9941
## noattrit$YearsInCurrentRole 2.683e-03 1.387e-02 0.193 0.8466
## noattrit$YearsSinceLastPromotion -7.644e-03 1.242e-02 -0.615 0.5384
## noattrit$YearsWithCurrManager -1.587e-02 1.442e-02 -1.101 0.2712
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.091 on 1188 degrees of freedom
## Multiple R-squared: 0.0395, Adjusted R-squared: 0.003926
## F-statistic: 1.11 on 44 and 1188 DF, p-value: 0.2889js.lm <- lm(noattrit$JobSatisfaction ~noattrit$OverTime+noattrit$MaritalStatus+noattrit$HourlyRate + noattrit$RelationshipSatisfaction)
summary(js.lm)##
## Call:
## lm(formula = noattrit$JobSatisfaction ~ noattrit$OverTime + noattrit$MaritalStatus +
## noattrit$HourlyRate + noattrit$RelationshipSatisfaction)
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.1355 -0.8005 0.1982 1.1236 1.4946
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.018010 0.143969 20.963 <2e-16 ***
## noattrit$OverTimeYes 0.132104 0.073547 1.796 0.0727 .
## noattrit$MaritalStatusMarried 0.037520 0.077789 0.482 0.6297
## noattrit$MaritalStatusSingle 0.191990 0.086500 2.220 0.0266 *
## noattrit$HourlyRate -0.003140 0.001524 -2.060 0.0396 *
## noattrit$RelationshipSatisfaction -0.049647 0.029088 -1.707 0.0881 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.089 on 1227 degrees of freedom
## Multiple R-squared: 0.01252, Adjusted R-squared: 0.008492
## F-statistic: 3.11 on 5 and 1227 DF, p-value: 0.00852#####Note: The fit is poor, but may provide directionality as to first order factors to assess.Linear Regression Modeling suggests that Overtime, Marital Status, Hourly Rate, and Relationship Satisfaction may be factors related to Job Satisfaction. This method didn’t work as well as with understanding of Job Roles. Only one of the four variables has a good correlation to Job Satisfaction, Hourly Rate.
Overtime
ggplot(data=as.data.frame(table(attrit$JobSatisfaction,attrit$OverTime,dnn=list("JobSatisfaction","OverTime"))), aes(x= reorder(JobSatisfaction,Freq),y=Freq, fill = OverTime)) +
geom_bar(stat="identity",position="fill") + coord_flip() + ggtitle("Percent of People Who Have Over Time is Consistent for All Job Satisfation Levels") + xlab("") + ylab("Percent of Employees") + geom_abline(slope=0, intercept=0.5, col = "black",lty=2)+theme(plot.title=element_text(hjust=0.5))+scale_fill_manual(values=alpha(c("steelblue4", "thistle4")))
Marital Status
ggplot(data=as.data.frame(table(attrit$JobSatisfaction,attrit$MaritalStatus,dnn=list("JobSatisfaction","MaritalStatus"))), aes(x= reorder(JobSatisfaction,Freq),y=Freq, fill = MaritalStatus)) +
geom_bar(stat="identity",position="fill") + coord_flip() + ggtitle("Job Satisfaction by Marital Status is Consistent") + xlab("Job Satisfaction Level") + ylab("Percent of Employees") + geom_abline(slope=0, intercept=0.5, col = "black",lty=2)+theme(plot.title=element_text(hjust=0.5))+scale_fill_manual(values=alpha(c("steelblue4", "thistle4", "paleturquoise4")))
Relationship Satisfaction
ggplot(data=as.data.frame(table(attrit$JobSatisfaction,attrit$RelationshipSatisfaction,dnn=list("JobSatisfaction","RelationshipSatisfaction"))), aes(x= reorder(JobSatisfaction,Freq),y=Freq, fill = factor(RelationshipSatisfaction))) +
geom_bar(stat="identity",position="fill") + coord_flip() + ggtitle("Job Satisfaction by Relationship Satisfaction is Consistent") + xlab("Job Satisfaction Level") + ylab("Percent of Employees") + geom_abline(slope=0, intercept=0.5, col = "black",lty=2)+theme(plot.title=element_text(hjust=0.5))+scale_fill_manual(values=alpha(c("steelblue4", "thistle4", "paleturquoise4", "mistyrose4")))
Find common demographics for people with high/lo job satisfaction
JS.4 <- noattrit[noattrit$JobSatisfaction==4,]
JS.3 <- noattrit[noattrit$JobSatisfaction==3,]
highjs <- rbind(JS.3, JS.4)
demohighjs<- highjs[1:8]
demohighjs$Attrition <- NULL
demohighjs$BusinessTravel <- NULL
demohighjs$OverTime <- NULL
demohighjs$MaritalStatus <- NULL
countdemohighjs<- count(demohighjs)
head(countdemohighjs[order(countdemohighjs$freq,decreasing = TRUE),],10)## Department EducationField Gender JobRole freq
## 21 Research & Development Life Sciences Male Research Scientist 53
## 60 Sales Life Sciences Male Sales Executive 39
## 17 Research & Development Life Sciences Male Laboratory Technician 38
## 33 Research & Development Medical Male Research Scientist 35
## 16 Research & Development Life Sciences Male Healthcare Representative 31
## 66 Sales Marketing Male Sales Executive 30
## 29 Research & Development Medical Male Laboratory Technician 29
## 15 Research & Development Life Sciences Female Research Scientist 28
## 57 Sales Life Sciences Female Sales Executive 25
## 63 Sales Marketing Female Sales Executive 24Visual inspection of the Top 10 common demographics suggests exoloring Department, Education Field, Gender and Job Role as factors related to Job Satisfaction
==
*(comment: Joblevel description was missing from the original dataset, so we only have the category number).